HOME | SEARCH | CONTACT | HARVARD LAW SCHOOL

[]

ABOUT PON
Negotiation Analysis Supplement NEWS
CLEARINGHOUSE

by Howard Raiffa

Welcome!

This Supplement presents an extension of the material in

the parent book, Negotiation Analysis (NA). When my
publisher balked at the size of the first draft of NA — then
known as Collaborative Decision Making — I was adamant
See Also:
that all must stay or else the picture would be incomplete.
About But I was torn: a book of 800 pages was a bit much. The
the Book
compromise solution, once imagined, was acceptable to
author and publisher: make all excess material including
mathematical appendices available on the Web. So here
you are and I am.

There is an added reason for this Supplement: adaptation

of the book plus Supplement for classroom use. Let me
refer to the book cum supplement as Negotiation
Analysis Supplemented and abbreviated as NAS. I did
not intend it this way, but I'm now of the opinion that NAS
when coupled with Smart Choices, a book written by myself
and two former Ph.D. students of mine, provides
surprisingly good coverage of the burgeoning new field
called Decision Science.

I had several venues for a Web site to choose amongst, but

the most appealing was through the auspices of the
Program on Negotiation, which I helped co-found with
Roger Fisher in the early '80s.

The separate parts of the Supplement are given here in PDF

format (and require a PDF Reader). Click the links below
each of the following synopses to download that piece of
the Supplement.

Table of Contents of the Supplement

A Detailed Table of Contents of NA.

This detailed Table of Contents takes several pages to
present and according to my publisher is redundant with a
detailed index, which is included in the book. My druthers,
space permitting, would have been for the inclusion of
both; but, happy compromise, NAS includes both.

PDF: 51 KB / 18 pages

NA Part Introductions and Chapter Summaries

PDF: 41 KB / 9 pages

My Short Vita
The Harvard University Press does not include much info
about its authors on the jacket cover. I would have liked
more. So, here is some more.

PDF: 5 KB / 1 page

The Role of a Neutral Joint Analyst

The Neutral Joint Analyst (NJA) is special type of
intervenor. At times he (or she or a team) acts as
facilitator, or as (proactive) mediator, or as (non-
evaluative) arbitrator; but the NJA is also part decision
analyst and game theorist who adds analytical skills in
helping the parties realize joint gains. The NJA may be an
invitee or invitor to the negotiation table. This short letter
from an NJA to possible negotiating parties describes what
the NJA has to offer. I'm intrigued with this intervenor.
Perhaps it's because I identify so openly with this character.

PDF: 12 KB / 3 pages

Chapter 10B: More on Auctions and Bids.

The material in Chapter 10 on Auctions and Bids is rather
extensive but stops at an inopportune place just when the
material gets really interesting. This Supplement continues
to examine the case when the prize is uncertain — like it
would be if the prize was a glass jar filled with coins and
where the bidding parties have different subjective
probability assessments about the value of this prize. It
also describes why in some cases the winner of certain
types of auctions tend to lose money overall — the so-
called "Winner's Curse." The Supplement also includes the
case where one of the bidding parties has privileged
information about the objectivre common value of the
prize. This is known in the literature as the case with an
"Unknown common prize with asymmetric information." It
illustrates the power of possessing good information: you
can intimidate other bidders.

The presentation is quite engaging, I would say, for those

of you who relish mathematical reasoning. Try it. Some,
with little math, have registered surprise at themselves:
they actually enjoyed struggling with more mathematics
than they thought they could tolerate.

PDF: 102 KB / 34 pages

An Appendix on Financial Risk Analysis.

It was not clear to me just where this material belonged.
Somewhere in Part III.

It examines the case where negotiating parties have an

opportunity to squeeze out joint gains by cleverly sharing
financial obligations and rewards — perhaps over time. The
parties may differ on probabilistic perceptions, on utilities —
e.g., on willingness to assume financial risks — and on the
time value of money. The parties could struggle to do this
by themselves, but how much more efficient this can be
done through the interventions of a neutral joint analyst.
The message rings clear: differences beget joint gains. The
topic has a vitality of its own and if the reader becomes
intrigued and wants more, then he would be well-advised to
read some of the seminal papers of Robert Wilson on
Syndicates and beyond.

PDF: 206 KB / 7 pages

The Two-Person, Zero-Sum Game

This could be an appendix to either Chapter four or ten. In
NA I spend quite some time on game theory but there is a
glaring void in the book on two important topics that were
historically central to the development and acceptance of
the theory: (1) the famous minimax theorem of the two-
person, zero-sum game, and (2) the power of bluffing. I try
in this and the next section to partially fill this void.

Consider the extreme case in which you are playing against

a strict adversary who has diametrically opposing interests
than you: if you prefer A over B, he, or she, prefers the
opposite. This is the essence of the zero-sum case. Now if
both parties have a finite number of strategies, then, you,
by playing cleverly, can guarantee a return of at least v and
your adversary, by playing cleverly, can guarantee you can
get at most v. The mathematics is profound and beautiful
and its inventor, John von Neuman, a genius. There is
another, non-recognized reason for the importance of this
type of security analysis. Consider the case where the row
player, say, knows his own payofs but knows precious little
about the payoffs of the other player,. The game cannot be
analyzed for equilibria because the players don't have
complete information. This happens quite often in practice
and doing security analysis on one's own payoffs may be a
starting point for further subjective analysis.

PDF: 51 KB / 12 pages

Security Analysis (as Done with EXCEL)

Shows how to find optimal, randomized strategies in a zero-
sum game by using a special, linear programming,
computer program, SOLVER, an add-in for EXCEL. In the
game context the game problem can be approached from
either the perspective of the maximizing row player or the
minimizing column player. These two formulations are said
to be dual to each other. The theory of linear programming
got a shot in the arm when von Neumann, on the basis of
the obvious duality of the special case of the zero-sum
game, posited a general theory of duality for all LP
problems. It has something to do with equilibrium pricing
and immediately LP became the darling of the mathematical
economist.

PDF: 26 KB / 5 pages

Simulations
 If the book is to be used as a text in a course, I strongly
recommend the instructor use several simulated negotiation
exercises to be suggested in the Supplement. Let me
explain how this might work.

For each of several chapters of the book, the materials in

the Supplement will suggest appropriate motivating
simulations. Suppose one of these simulations involve two
parties — say labor and management — and you are
assigned to play the role of labor; you will then be paired
with a management player and arrange for negotiation
time. Both of you will be expected to read a couple of pages
or so of general information and, given your role
assignment, you will have to read and study some
confidential information for "labor eyes only." You then
negotiate with your management counterpart in either face-
to-face negotiations ( the most preferred) or by telephone
or by e-mail and keep records of what occurred. The
controller of this simulation collects the data from many
labor/management pairs and prepares a briefing report
giving a statistical analysis of what worked and what did
not. As in duplicate bridge, you should now ponder how well
you did versus how well other labor players did when
playing the identical simulation. Now you are ready to read
the appropriate chapter in NA covering some of this
material in a more active manner.

Download the Entire Current Version of the

Supplement
This file includes the preceding Table of Contents

PDF: 93 pages

Problems downloading or opening these documents? Please

contact Ed Hillis at [email protected].

(Modifed June 27, 2003)

Copyright © 2003 The President and Fellows of Harvard College.

Negotiation Analysis: The Science and Art of Collaborative
Decision Making by Howard Raiffa (with contributing authors
David Metcalfe and John Richardson), Harvard University
Press, 2003.

DETAILED TABLE of CONTENTS

Preface

Part I: FUNDAMENTALS
Chapter 1: Decision Perspectives
Our Approaches to Decision Making
Decision Analysis
Behavioral Decision Theory
Game Theory
Negotiations
Individual Decisions:
Prototypical Examples
Drill or Not Drill
Surgery or Radiation
Invest or Not
Group Decisions
Interactive decision making: Theory of Games
Joint decision makng: Negotiation theory
Fuzzy boundaries
Descriptive, Normative, and Prescriptive Orientations
The Individual Case
Descriptive decision making: How decisions are mde
Normative decision making: How decisions should be made
Prescriptive decision making: How decisions could be made better
The case of transitivity
Group Orientations
Symmetrically descriptive
Symmetrically normative
Asymmetrically prescriptive/descriptive
Descriptive, normative, and prescriptive orientations for the external helper
Core Concepts

Chapter 2: Decision Analysis

Generic Unitary Decisions in Negotiation Contexts
The PrOACT WAY of THOUGHT
Identifying the Problem
Clarifying Objectives
“Objectives” = “Interests”

Copyright  2003 Howard Raiffa

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Detailed Table of Contents

Why and What

Generating Creative Alternatives
How?
The More the Better
Evaluating the Consequences
Conditional Analysis
Dominance
Making Tradeoffs
Costing out
Equal Swaps
Uncertainty and Risk
Defining Risk Profiles
The separation principle
Value analysis
Uncertainty analysis
Ordinal Ranking
Expected Monetary Values (EMV)
Choice between lotteries with monetary payoffs
Limitations of EMV
Expected Desirability Values (EDV)
Intensity of Feeling
The Argument for EDV
Limitations for EDV
Expected Utility Values (EUV)
Risk Profiles with BRLT consequences
Monotonicity, continuity, substitutability
Change of Scales
Core Concepts

Chapter 3: Behavioral Decision Theory

Decision Traps
The Anchoring Trap: First Impressions
The Status Quo Trap: Sticking with the Past
The Sunk Cost Trap: Trying to Recoup Losses
The Confirming Evidence Trap: Seeing What You Want to See
The Framing Trap: Solving the Wrong Problem
Prescriptive Advice
Uncertainty Anomalies
On Thinking Probabilistically
Conditionality Ambiguities
The Monty Hall Conundra: Brain Teasers about Conditional Probabilities
The Overconfidence Trap
The Conjunction Fallacy
A bit of probability theory
The Base Rate Fallacy
Underestimating the Value of Sample Evidence
Getting Mystical about Coincidences
Choice under Uncertainty
The Allais Paradox
The Descriptive Attack against EUV: Empirical Results

Copyright  2003 Howard Raiffa 2

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Detailed Table of Contents

The Normative Defense of EUV

The Sure-Thing Principle
The Ellsberg Paradox
The Descriptive Attack against Use of Subjective Probabilities: Vagueness
The Normative Defense of Subjective Probabilities
Core Concepts

Chapter 4: Game Theory

Game Theory and Negotiation Analysis
The Strategic Essence of Interactive Thinking
Two-by-Two Matrix Games
The Rules of the Game
Fixed Strategies
Two Alternatives
Common Knowledge
Simultaneous Choice
Perfect Information
No Cheap Talk
Consequences and the Matrix Display
Game #1: Indeterminacy
Game #2: Dominance
Game #3: Iterative Dominance
Game #4: Equilibria
Choice of Equilibria
Game #5: Asymmetric Prescriptive/Descriptive Analysis
Empirical Results
Game #6: An Advantage in Going First
Game#7: Threats with Pre-play Communication
Game #8: Doomed by Rationality
Social Traps
Prisoner’s Dilemma
Game #9: Chicken and the Temptation of Mutual Annihilation
Game #10: The Iterated Dilemma Game and Tacit Collusion
Tacit Collusion
The Importance of Trust
The Fragility of Trust
Operational Trust without Empathetic Trust
The End Game
Equilibria Analysis
Advice on How to Play:
Start cooperatively
Be provocable
Don’t be vindictive – use measured punishment
Game # 11: The Ultimatum Game and the Rewards of Irrationality (?)
Joint Normative Equilibria Analysis
Asymmetric Prescriptive/Descriptive Analysis

Core Concepts

Copyright  2003 Howard Raiffa 3

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Detailed Table of Contents

Chapter 5: Negotiation Analysis

The Essence of Joint Decision Making
Joint Decisions
Joint Payoffs
Reciprocal Communication
Creativity
Decision Perspectives for Negotiators
Individual
Interactive
Joint
The Negotiator’s Dilemma
Creating and Claiming
Full, Open, Truthful Exchange (FOTE)
Collaborative Decision-Making: Real-World Examples
The Importance of a Polar Extreme
Partial, Open, Truthful Exchange (POTE)
Some Organizing Questions
Dispute or Deal?
Are There More than Two Parties?
Are the Parties Monolithic?
Is the Game Repetitive?
Are There Linkage Effects?
Is There More than One Issue?
Is an Agreement Required?
Is Ratification Required?
Are Threats Possible?
Are There Time Constraints or Time-Related Costs?
Are Contracts Binding?
Are Negotiations Private or Public?
What Are the Group Norms?
Cooperative Antagonists
Strident Antagonists
Collaborative Partners
Can an External Helper Be Used?
Core Concepts

PART II: TWO-PARTY DISTRIBUTIVE NEGOTIATIONS

Chapter 6: Elmtree House

The Problem — and the Opportunity
Preparing for Negotiations
Gathering Information
Preparing a strategy for negotiations
Opening Gambits
The Dance of Concessions
Closing Gambits

Copyright  2003 Howard Raiffa 4

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Detailed Table of Contents

Core Concepts

Chapter 7: Distributive Bargaining

The Distributive Abstraction
Reservation Values (RVs)
Zone of Possible Agreement (ZOPA)
Buyer's and seller's surpluses
The geometry of distributive bargaining
The Streaker: A Simple Role-Playing Exercise
The dance of concessions
Empirical results
Variations
ANALYTICAL ELABORATIONS
Case 1. Each party knows the other’s reservation price
Case 2. One party knows the other’s reservation price
Case 3. Common probabilistic information: Game theoretic analysis.
Informal bargaining surprisingly efficient
Double Auction: Simultaneous Revelation.
Game theoretic equilibrium analysis
People Outsmart themselves
The more they lie, the more it pays to be honest
… and vice versa.
Spreading idealistic, collaborative behavior
Core Concepts

Chapter 8 : The Complication of Uncertainty

CASE STUDY: THE SORENSEN CHEVROLET FILE

Background of the Case
Decision Analysis by the Defendant
Decision Analysis by the Defendant’s Insurance Company, UGI
The Plaintiff’s Perspective:
Deborah’s Decision Tree
Being Specific about Uncertainties
Deborah’s Strength of Preference Analysis
Debra’s Utility Analysis
GENERALIZING: DEALING WITH UNCERTAINTY AND RISK
Speaking the Language of Probability
How Real People Do Act
Core Concepts

Chapter 9: The Complications of Time

The Timing of Concessions

Copyright  2003 Howard Raiffa 5

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Detailed Table of Contents

Sequential Search
Select the Most Beautiful Woman Problem
The Strike Game
The Simulated Strike Game
Empirical Results
The Escalation Game
The Problem:
How Real Players Behave
Prescriptive Analysis
Behavioral Insights
The No-Strike Strike or the Virtual Strike
Escrow Accounts and Penalties
Creation of Strike Mechanisms
The New York State Budget Deliberation
Core Concepts

Chapter 10 Auctions and Bids

Intertwined Negotiations
Distributive Bargaining with One Seller and Many Buyers
AUCTION MECHANISMS
Open Descending Outcry Auction - Dutch Auction
A Bidder’s Decision Analysis
Competitive Sealed Bids
The General Case: Private Valuations and Fuzzy Knowledge.
MBOO Analysis
The Philatelist or Vickerey Auction (High Bidder Wins At Second-High Price)
The Auctioneer’s Perspective.
SPECIAL CASES
Common Knowledge and Common Values
Bidding For $100 Against One Other Bidder
Uncertain Common Value, Known Objective Probabilities, Unknown Private Risk Aversions
Reciprocal Buy/Sell Bids
Combinatorial Bidding - The FCC Spectrum Auction
Privatization of Governmental Monopolies
Serial Bidding
Parallel Bidding
Design of the FCC Auction
FOTE Analysis
Core Concepts

PART III: TWO-PARTY INTEGRATIVE NEGOTIATIONS

Chapter 11: Preparing for Negotiations

SETTING THE STAGE

Identifying a Potential Deal; Early Contact

Copyright  2003 Howard Raiffa 6

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Detailed Table of Contents

Three Phases of Preparation

Phase 1. Preparing alone
Phase 2. Preparing together
Phase 3. Preparing alone
PHASE 1: PREPARING ALONE
Examining Interests (Objectives, Fears, Concerns, Wish Lists)
Envisioning the Future
Cataloguing Resources
Comparative advantages
Exploring Alternatives to Agreement
Generating Options for Agreement
Visioning the Future
Assessing Uncertainties
Gathering Information on the Other Side
Exploiting Differences
Invoking Objective Criteria
PHASE 2: PREPARING TOGETHER
Dialoguing or Pre-Negotiating
Planning the Logistics
Setting Goals for the Meeting
Creating the Ambience
Sharing Interests
Discussing and Agreeing on Process
Preparing the Template
Identifying the Issues to Be Resolved?
Specifying Possible Resolutions of Each Issue?
Delegating Template Construction to a Joint Sub-Committee

CASE STUDY. AMPO vs. CITY

Core Concepts

Chapter 12: Template Evaluation

Tradeoffs, BATNAs, and RVs
Template Construction and Template Evaluation
Preferences and Value Tradeoffs
Within-Issue Qualitative Analysis
Between-Issue Qualitative Analysis
Within-Issue Quantitative Analysis
Between-Issue Quantitative Analysis
The Additive Scoring System
Appropriateness of an Additive Scoring System
Cutting Corners
BATNAs and Reservation Values
Creating Best Alternatives
Linking Best Alternatives With Reservation Values
Cutting Corners and Practicalities

Copyright  2003 Howard Raiffa 7

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Detailed Table of Contents

Monetary Quantification
Core Concepts

Chapter 13: Efficiency, Equity and Fair Division

Efficient Contracts
Joint Ordinal Rankings
Choice of a Contract
Efficiency: Dominance and Elimination of Non-Contenders
Joint Cardinal Rankings
Randomization and Extreme Efficiency
Extreme Efficiency and the Ruler Representation
FAIR DIVISION
The Problem
Some Proposed Solutions
Sequential Choice
Toss a Coin
Divide and Choose
Allocation of Points (with item to highest bidder)
FOTE Analysis
Ratios of Scores and Characterization of Extreme-Efficient Points
What is Fair?
Inadequacies of Other Solutions
Choosing Sequentially
Divide And Choose
The Lesson of Experience
FOTE ANALYSIS WITH SPREADSHEETS (OPTIONAL READING)
Organizing the Spreadsheet
DesirablePayoffs
Maximizing the Sum of Marty and Janet’s Scores
Maximizing The Minimum:
Pure and Mixed Strategies
Comparison Of Procedures

Core Concepts

Chapter 14. JOINT TEMPLATE ANALYSIS (The General Case)

NELSON VS. AMSTORE - A CASE STUDY

The Scored Template
Compilation of All Contracts
Graph of All Joint Evaluations
Decomposition for Large Problems
Efficient Contracts
Extreme-Efficient Contracts
Use of C-Weights
Incremental Analysis
Numerical Illustration

Copyright  2003 Howard Raiffa 8

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Detailed Table of Contents

Analysis by Critical Ratios

Graph of Extreme-Efficient Contracts
BATNAs and Reservation Values
ANALYSIS WITH SOLVER
The Programming Problem
Legitimacy and Feasibility
Maximum Feasibility
Using SOLVER to Find Equitable Efficient Contracts
Maximizing: Sum of Payoffs
Product of Payoffs
Minimum POP
Comparison of Solutions
Core Concepts

Chapter 15: Real Behavior in Negotiations

LABORATORY AND CLASSROOM SIMULATIONS

Scorable Negotiations
Debriefing
Wide Dispersal of Results
Joint Gains Left on the Table
Negotiating an Agreement
The Dance of Packages
Joint Construction of a Compromise Package
Getting started
Constructing a Compromise
Closing the contract
BEHAVIORAL REALITIES IN NEGOTIATIONS
Interactive Biases, Anomalies, and Errors
Zero-Sum Bias
Social Utility
Reactive Devaluation
Negative Attributions
CULTURAL DIFFERENCES IN NEGOTIATIONS
Differences Within and Between
Joint membership
Differences in Interests
Differences in Negotiation Style
The Us/Them Stance

Core Concepts

Chapter 16: Negotiating without Reciprocity

The Asymmetric Prescriptive/Descriptive Approach
Negotiating against Diverse types
The Simple Cooperatiive
The Aggressive Claimer

Copyright  2003 Howard Raiffa 9

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Detailed Table of Contents

The Disputatious Adversary.

Negotiating Against a Simple Cooperator
Characteristics of the Simple Cooperator
Creating Value with a Simple Cooperator
The Principle of Tentative Agreement
Eliminating and Rearranging Resolutions: within Issue Qualitative Efficiency
Analysis
Sharing Information about Qualitative Importance of Issues
Assignments on Issues Exploiting Imbalances of Importance Evaluations
Claiming Value with a Simple Cooperator
RVs Without a Scorable Template
Thinking about Our Aspiration Level
Positional Bargaining with the Aid of a Template
Handling Deadlock
A Process De-emphasizing Claiming
The Double Auction Method
Negotiating Against an Aggressive Claimer
Preparation
Phase 1 Preparation: Anticipating a Change from FOTE to POTE
Phase 2 Preparations: Prenegotiations
Fostering Collaboration: Moving Up the Ladder
Emphasize Future Interactions
Insecure Contracts: Trust for the Untrustworthy
Negotiating Against a Disputatious Adversary
The Essence of Disputes
Case Study of a Business Dispute between Antagonistic Partners
Creative Preparation Alone
Focus on Process
Anticipating Disputatious Behavior
Threat and Counter-Threat
Irreversible Commitments
Changing the Geometry
Do You Want to Play?
Use of Agents
Use of a Third-Party Helper

Core Concepts

PART IV: INTERVENTIONS IN TWO-PARTY DISPUTES

Chapter 17: Mostly Facilitation and Mediation

The Need for Help

Types of Problems
Quarreling siblings with mom as H
The manager as H
ADR for contract disputes
Collaborative lawyering
Divorce or family disputes
Court mandated, non-binding arbitration

Copyright  2003 Howard Raiffa 10

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Detailed Table of Contents

The judge as mediator

H initiates proceeding
The Vatican intervenes in a territorial dispute
Camp David
Valleta Accord for the CSCE
Track II Negotiations
The psychiatrist as mediator
Types of Help
Disputes with Heavy Interpersonal Conflicts
Deals with Non-Collaborative Protagonists
Roles of an External Helper
Roles of an Analytical External Helper
Negotiating the Mandate for the External Helper
Interests of the External Helper
Just another Negotiator?
Why External Helpers Are Not Used
Case Study: Camp David
Historical Background
Preparing the Mediator: The US Team’s Approach
Interests of Mediator
Single Negotiating Texts
Inducements Offered by Mediator
Critique of SNT Process
Core Concepts

Chapter 18 : Mostly Arbitration

EVALUATIVE INTERVENTIONS
Types of Arbitration
Conventional Arbitration
Bounded Arbitration
Selective Arbitration
Final Offer Arbitration
Baseball Salary Negotiations
Rationale
NON-EVALUATIVE INTERVENTIONS
The Neutral Joint Analyst
Combining POTE with a double auction mechanism
Post-Settlement Settlements: Embellishment
Embellishment in the absence of agreement
Putting it into practice
Division of Mrs. K’s Art Collection
APPENDIX: GAME THEORETIC ANALYSIS OF FINAL OFFER ARBITRATION
All’s Not Well with the Procedure
Analysis of Special Cases
Value of the Ideal Is Known
Commonly Perceived Probability Distribution for the Ideal
Differing Probability Distributions for the Ideal

Copyright  2003 Howard Raiffa 11

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Detailed Table of Contents

Chapter 19: WHAT IS FAIR?

Canonical Representation of the Problem of Equity
The Nash Bargaining Problem
Units of Measurement
Fairness Proposals
Equity Analysis: Setting Up the Problem
The Maximin or Equipop Point
The Mid-Mid Point
The Maimonides Solution
The Nash Solution
An Example: Never on Thursday (or Oot vs. Tath)
Rationale for the Nash Solution
Numerical Illustration
Independence of Irrelevant Alternatives
The menu interpretation
Questioning the Universality of the IIA Principle
The Balanced- Increment Solution (The Raiffa Solution)
Back to Units of Measurement and Utilities
Rich Man/Poor Man Dilemma
Core Concepts

Chapter 20: PARALLEL NEGOTIATIONS

ORIENTATION
Types of Parallel Negotiations
Front Channel
Back Channel
Track II Negotiations
Blurred Boundaries
A Hypothetical, Abstract Problem of Intractable Negotiations
Barriers to Negotiation
COPING WITH THE BARRIERS
High Political Cost of Initiating Unproductive Negotiations
“Experimental Non-Negotiations”
Troublesome Logistics
Posturing, Positional Bargaining, and Excessive Claiming
Inability to Brainstorm
Inability to Use FOTE Collaboratively
Front-Channel Negotiators Are Not Natural Compromisers
Power Imbalances
Lack of Trust and Insecure Contracts
Securing Insecure Contracts
Ripeness of the Conflict
CSCE Dispute Settlement Mechanism: The Valletta Accord
Divided Selves: Blocking Coalitions of Extremists
Ignoring Future Joint Collaborative Opportunities
The BCP Project.

Copyright  2003 Howard Raiffa 12

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Detailed Table of Contents

Option Value of Future Collaborative Opportunities.

HOW IT IS (AND COULD BE) DONE

Logistics of Getting to the Table in Track II Negotiations
Pre-Mediation Briefing Reports and Track II Shuttle Analysis
Case Study: Peru-Ecuador Conflict 1995
The Conflict
Facilitated Joint Brainstorming (FJB)
After Effects
Core Concepts

PART V: MANY PARTIES

Chapter 21: Collaboration with Many Actors

Introduction and Motivation
Types of Groups
Why Groups Do Better than Individuals
The Pooling of Knowledge
The Inclusion of Expertise (e.g., legal, financial, ability to draft documents,)
The Possibility of Parallel Tasking of Work
The Decomposition of Tasks to Exploit Knowledge and Expertise
The Pooling of Judgment (e.g., average is better than individual assessments of uncertainties)
The Protection against Bias and Fraud
The Implementation of Ideas (e.g., more advocates)
The Inclusion of Influentials to Help The Selling of Policy to Others
The Desire for within Group, Social Interactions
The Benefits of Democratic Representation
The Courage to Advocate Unpopular Views
The Lack of a Single Synthesizing Mind
An individual can intuit – a group can’t
A group has no synthesizing mind and therefore it must be more systematic
BEHAVIORAL REALITIES (Problems with Group Behavior)
Why Groups Do Worse than Individuals
Communications Overload
People all talk at the same time.
No one can make himself heard, and worse, can’t hear others.
People don’t listen carefully. They think about their next input.
Coordination Loss
They forget about what was said; and no record is kept.
Discussions are disorganized, go around in circles, and easily get sidetracked.
They pull in different directions.
Cognitive Overload
Disengagement
Some parties disengage from interactions. Some withdraw. Some free ride.
No one takes responsibility
Responsibility too diffuse
Interpersonal Conflict
Group Think
Coalition Dynamics

Copyright  2003 Howard Raiffa 13

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Detailed Table of Contents

PRESCRIPTIVE SUGGESTIONS
Membership
Invite only those who contribute
Wanted: Ideas and expertise
Managers of process (facilitating, drafting)
Help in getting approval and in implementation
Organization of the Substance
Choosing purpose
Early articulation
Need for tangible end-product
Nested aspirations
Basic questions (Why us? Why now? ….)
Structure
No group mind for synthesis
Need for decision- making framework
PrOACT
Other problem-solving frameworks: circle chart; seven elements;
Ben Franklin's pros and cons
Using Labor Efficiently: Organization of Tasks and Sub-Committees
Decomposition of Tasks
Assignments to Sub-Committees
Recomposition and Synthesis
Style: Organizing the Conduct of Meetings
Facilitation
Documentation
Brainstorming
Time
INDIVIDUAL DECISION MAKING WITH EXPERTS AND ADVISORS
The Advocacy Model
The Structured Analysis Model

Chapter 22: CONSENSUS

Consensual Agreement
FAIR DIVISION WITH MANY PARTIES
Analysis with Monetary Transfers
Naive procedure
Auction procedure
The Steinhaus-Knaster Allocation
Strategic Misrepresentations with the Steinhaus Procedure
Analysis without Monetary Transfers
Three Parties: The Problem and Its Analysis
Comparison of Solutions
A CONRETE, BUT ABSTRACT, CASE
FOTE Analysis with SOLVER
Comparison of Solutions
Disengaging from FOTE Analysis: POTE Analysis
Negotiation Template Used as Decision Aid

Copyright  2003 Howard Raiffa 14

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Detailed Table of Contents

Friends and Enemies

Diagram of Affinity and Opposition

Voting Dynamics
First Vote
Second Vote
Third Vote
Scores on Votes
The Two-Party Counterpart of the Voting Procedure
Comparison of Solutions
Summary
Strategic Claiming Behavior
The $64,000 Question

Core Concepts

Appendix: Consensus with Asymmetric Claims

Chapter 23 : COALITIONS
Coalition Formation
Cooperative Game Theory
The Characteristic Function Form
Case Study: The Scandinavian Cement Company
The Core
A Pure Coalition Game
Game Description
Offers that Cannot Be Readily Refused
Face-to-Face versus Terminal-to-Terminal
Rationality, Fairness and Arbitration
The Core of the Game
The Shapley Value
A Modification of the Shapely Values
The Game with One Strong Player
Moving Toward Reality
Intertwining Negotiations
Core Concepts

Chapter 24: Voting

Majority Rule
WYZARD, Inc.
Independence of Irrelevant Alternatives
Insincere Voting
Arrow’s Impossibility Theorem
The Problem
Strengths of Preferences

Copyright  2003 Howard Raiffa 15

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Detailed Table of Contents

Strategic Voting
Strategic Ordering of Bills in a Legislature
Insincere Voting Revisited
Randomization
Strength of Preference and Logrolling
Implications for Many-Party Negotiations
A Real Case: The Voyager Mission To Outer Planets
The Collective Choice Problem
Collective Choice Procedures and the Independence of Irrelevant Alternatives
Postscript
Core Concepts

Chapter 25: Pluralistic Parties

Introduction
Agents with Multiple Principals
An Agent with Two Principals: A Hypothetical, Simple Case Study
The Problem
How Negotiations Are Conducted
How Negotiations Should Have Been Conducted
Internal Differences
Internal Payoffs (by Formula)
The Agent’s Optimal Mandate: Internal Compensation
Use of SOLVER
Synchronization of Internal and External Negotiations
Creative Internal Compensation
Blocking Coalitions
Winners Compensating Losers
Within Contract Modifications
Adding New Issues
Using New Negotiations with Same and Different Parties
Internal Side Payments
Logrolling Internally
The Panama Canal Negotiations
Prelude to Negotiations
Initial Negotiations and Internal Preparations
Course of Negotiations
Internal Conflict
Structure
Assessment of uncertainties
Conditional evaluations and tradeoff analysis
Synthesis for choice
Fair Division with Three Parties*
Fair Division with Possible Side Payments
Core Concepts

Copyright  2003 Howard Raiffa 16

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Detailed Table of Contents

26: Interventions in Multiparty Negotiations

Introduction
The Problem of Complexity
Roles of an Intervenor
Hypothetical Case Study: Interventions in a Multi-Party Community Dispute
The Problem
Introduction of the facilitation/ mediation Team
Interests of the intervenor
Protocols for helping each side
Negotiating role of the intervenor
Ericson Farm Planning Committee
Interests of the town
Tasks and sub-committees
Options for negotiations
Public Hearings and the Ratification Process
Competing Plans for the Town Land and a Voting Process
The Ericson Charitable Offer
The Denouement

ENERGY NEGOTIATIONS IN THE EUROPEAN UNION

The European Commission’s 1988 Policy Paper:
The Conflicts of Interest
The Irish Presidency of the Council (July-December 1996):
The Nine Issues to be Resolved
The Netherlands Presidency of the Council (January-June 1997):
Starting SNT Is Not Feasible Nor Amendable
The Luxenbourg Presidency of the Council (July-December 1997)
Sharing Information about Interests with a Template as a Communication Aid
Finding a Feasible Contract
Applying FOTE Analysis to the Natural Gas Negotiations
The (Imputed) Scored Template and Reservation Values
Use of SOLVER for POP Analysis of the Netherland
and Luxenbourg Proposals
Finding the Nash and Maximin POP Solutions.
Core Concepts

Chapter 27: Social Dilemmas

The Many Party Social Dilemma Game

The Basic Game: No Pre-Play Communication, Private Choice
Variants
Pre-play communication, private choice
Pre-play communication, public choice
Repeated play
COMMONS PROBLEMS
The Mathematics of Sheep Grazing
A monopolist’s analysis

Copyright  2003 Howard Raiffa 17

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Detailed Table of Contents

The competitive analysis

Equilibrium analysis
Entry
On Not Catching Whales
Simple Abstractions
A Monopolist’s Analysis
Competitive Analysis
Free Entry
Joint Management
Need for Central Authority
The Case of Global Warming
Population control
The LULU Problem
LULUs, NIMBY, BYBYTIM
The Dutch Auction
Environmental Justice
Creative Compensation
PERSONAL ETHICS AND THE SOCIAL DILEMMA
The Mattress Problem
Free-Riders
Will the Hero Please Step Forward
Personal Charity
Some Personal Reflections
Core Concepts

CONTENTS
BIBLIOGRAPHY

Copyright  2003 Howard Raiffa 18

Negotiation Analysis: The Science and Art of Collaborative
Decision Making by Howard Raiffa (with contributing authors
David Metcalfe and John Richardson), Harvard University
Press, 2003.

Part Introductions

Part I: Decision-Making Perspectives

Part II: Two-Party Distributive Negotiations
Part III: Two-Party Integrative Negotiations
Part IV: Interventions in Two-Party Disputes
Part V: Many-Party Negotiations

Part I: Decision-Making Perspectives

Part I comprising chapters one to five is an extended introduction to the book

which develops analysis for negotiation -- not analysis of negotiation -- drawing
extensively on:
• Decision Analysis -- individual prescriptive decision making
• Behavioral Decision Theory -- judgmental decision making
• Game Theory -- interactive decision making.

Chapter one contrasts three perspectives of individual decision making:

• Normative analysis -- how ideally decisions should be made (by super rational
individuals without psyches);
• Descriptive analysis -- how real people actually do behave (very often at
variance with the normative abstraction);
• Prescriptive Analysis -- how real people could behave better (more
advantageously) with some systematic reflection.

In group decision making this threefold breakdown becomes more complicated.

Much of the book gives partisan (prescriptive) advice to one decision maker in a group
assuming the other individuals are not privy to that advice and where the others behave in
a more descriptive fashion. In contrast to this orientation, the theory of games takes a
jointly normative approach.

Chapter 2 is a tutorial that develops enough of decision analysis to be of use in the

rest of the book, but it leaves for later, just-in-time development of additional topics.

Copyright  2003 Howard Raiffa

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Full Overview

Chapter 3 examines real behavior and indicates that many individuals, some of
the time, do not act in conformity with rational behavior -- especially when uncertainties
loom large. We call such examples of "deviant" behavior anomalies, biases, errors, or
traps and we try to rationalize why and how they might occur.

Chapter 4 develops many of the concepts of the non-cooperative theory of games

in a tutorial, self-discovery, pedagogical style; but as in chapter 2 it leaves special topics
in the theory for just-in-time, later development where needed.

Chapter 5 provides an outline of the rest of the book. It introduces the concept of
idealized, joint behavior in which all protagonists in a negotiation agree to negotiate in a
truly collaborative manner agreeing to tell the truth and all of the truth. We call this joint
FOTE analysis -- Full, Open, Truthful Exchange. We claim that often in actual deal
making in contrast to dispute settling that this ideal is often approximated and the FOTE
examination sets up an ideal against which other approaches can be compared. In the
sequel, we back away from FOTE analyses to consider POTE-like behavior in which
some protagonists tell the truth but not the whole truth -- Partial, Open Truthful
Exchange -- leaving their bottom lines hidden from view of others; and back away further
to NOTE behavior -- No Open, Truthful Exchange.

There is a tension in negotiations between creating actions designed to build a

bigger pie and claiming actions designed to get a large share of the pie so created. The
main concern of this book is involved in the creating function but the claiming function
cannot be ignored. Part II of the book (chapters 6 -10) is mostly about claiming behavior
for two-party negotiations; Part III (chapters 11- 17) mostly about creating joint gains;
Part IV (chapters 18-20) about external interventions -- how facilitators, mediators, and
arbitrators could help; and Part V (chapters 21-27) complicates all by introducing more
parties into the fray.

PART II: Two-Party Distributive Negotiations (Chapters 6-10)

Chapter six presents a case study (Elmtree House) of the prototypical negotiation
problem of this part of the book. A seller (in this case an institution that owns a halfway
house) wishes to sell an asset (its residence, the house) to a buyer (a developer). The
seller wants more; the buyer less; and they haggle. It’s all a matter of claiming a larger
part of a fixed size pie. There are two parties that negotiate over one issue: money – but it
could be time or any other single commodity.

In the next chapter the problem is abstracted and the discussion addresses such
questions as: How to prepare? Who should declare first? Why? How extreme a first
offer? What’s a reasonable counter-offer? What is the pattern of concessions? And so
forth. The chapter discusses a related double-auction game in which the seller and buyer
simultaneously offer sealed bids and a transaction takes place if and only if the bids are
compatible. This game is analyzed normatively, descriptively and prescriptively and we
conclude that good old-fashioned haggling might be better.

Copyright  2003 Howard Raiffa 2

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Full Overview

Chapter eight complicates the picture by introducing uncertainty, like the

uncertainty of a trial outcome. We analyze the case of negotiating an out-of-court
settlement in a liability case. The dominating uncertainties are who would win the court
battle and the size of the jury award if the plaintiff prevails. We examine the problem as
an individual decision problem under uncertainty and illustrate and extend the
methodology introduced in Chapter two.

Chapter nine introduces the complication of time. It starts off by a seller

dynamically searching for buyers. When to quit searching is another one of those
decision problems under uncertainty. The central case of this chapter is the discussion of
a hypothetical strike game and this is analyzed from the perspective of the individual
decision maker and of game theory. Empirical laboratory observations demonstrate the
potential for dysfunctional escalatory, vindictive behavior. Strike mechanisms have their
problems but sometimes negotiating parties may wish there were a strike mechanism.
Well that’s an opportunity for negotiation.

An Appendix to Chapter 9, presented on the Web, considers risk sharing and

contingent contracting. We illustrate a case where the parties engage in a distributive
bargaining problem with no zone of possible agreement. The negotiating parties,
however, may disagree on probabilistic assessments, on attitudes towards risk and on
time value of money (i.e., different discount factors) and these differences can be
exploited in seeking joint gains through contingent contracting and risk sharing. There
are now possibilities of creating a bigger pie. But still the problem is mainly one of
claiming a larger share of the pie. This appendix serves as segue into Part III of the book.

Chapter ten considers the case where there are many buyers and just one seller.
The seller could negotiate sequentially with each potential buyer or engage in a
competitive bidding or auction procedure. The chapter compares several different types
of auction and competitive bidding procedures and we use individual decision analysis to
give partisan prescriptive advice to one of the players. We end the chapter by considering
the all-too-common case when two partners want to dissolve their partnership by having
one partner buy out the other.

We supplement the chapter by considering some advanced topics on the Web:

such problems as bidding for the rights for a commodity whose worth is very uncertain –
like bidding for rights to drill for oil. Empirically, the winners of such auctions end up by
losing money -- the so-called winner’s curse. Why is this so? Life becomes even more
complicated when there are asymmetries in knowledge about the uncertainty in question
– like when one party has confidential exploratory information about the potential size of
an oil field.

Part III: Two-Party Integrative Negotiations (Chapters 11-16)

This part of the book, comprising chapters 11 to 16, deals with Two-Party
Integrative Negotiations. By "integrative negotiations" we mean: negotiations having the
potential of resulting in joint gains. Part II dealt with Two-Party Distributive Negotiations

Copyright  2003 Howard Raiffa 3

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Full Overview

which involved partitioning a fixed-sized pie. Whereas Part II was mostly about claiming
tactics, Part III will be mostly about creating tactics -- how to create a bigger pie. But
there is a tension between tactics used to create a larger pie and tactics used to claim a
large portion of the pie created. How to balance this tension is part of the art and science
of negotiation.

Much of negotiation involves the settlement of disputes. The bulk of our attention
in Part III, however, will be with examining deals as presenting opportunities for joint
gains (accompanied, of course, by some distributional strains) in contrast to disputes or
problems that have to be resolved. Behaviorally speaking, negotiations involved in deal-
making tend to be more collaborative than those involving the settlement of festering
disputes. But still, much of the advice given for deals are relevant also for disputes. Part
of this advice is to try to convert a dispute into a deal and to prevent a deal from
becoming a dispute.

Chapters 11 and 12 concentrate on preparing for negotiations after the potential

for such deal-making has already been flagged. We don't examine how we might find
promising potential deals. We propose three phases in preparing for negotiations. In
Phase 1, the parties deliberate separately and alone and consider their interests,
objectives, visions; their alternatives to the present negotiation under review; their
options; and gather information about the other side about their interests, alternatives and
about their negotiation veracity and style. In Phase 2, the parties engage in an informal
dialogue in which they feel each other out; they selectively and adaptively share
information; they brainstorm together; they try to establish an appropriate relationship;
and they decide whether they should proceed further. If they see a green light ahead, we
suggest that they jointly decide just what must be decided. We strongly suggest that they
construct a template to guide their further negotiations. The template specifies the issues
that have to be decided and possible resolutions for each issue. In Phase 3 of their
preparation, they individually and confidentially evaluate the template by clarifying their
preferences for different contract outcomes and by exploring their Best Alternative To a
Negotiating Agreement (BATNA). Some analytically-minded party might go so far as
quantitatively scoring the template and establishing a numerical Reservation Value (RV)
which specifies the minimum value that would be acceptable in its negotiations. Chapter
11 deals with Template Construction, 12 with Template Evaluation, and Chapters 13and
14 with Template Analysis. Chapters 11 to 14 have a jointly normative flavor; 15 is
mostly jointly descriptive, 16 mostly asymmetric prescriptive/ descriptive -- i.e., we give
partisan prescriptive advice to Party A based on a descriptive analysis of the behavior of
Party B.

Chapter 13 examines an idealization: both parties negotiate agreements using

Full, Open, Truthful Exchange (FOTE) -- they tell each other the truth and the whole
truth. [It was a surprise to me to find that in real-world, deal-making negotiations, this
ideal is often the reality.] The chapter discusses efficient and equitable contract outcomes
for the special case in which two parties have to share a set of indivisible items, the so-
called Fair Division Problem. The problem is special in two ways: (1) the ith issue of the
template involves what should be done with the ith item and there are just two possible

Copyright  2003 Howard Raiffa 4

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Full Overview

resolutions (give it to A or to B -- no sharing allowed); and (2) there is no alternative to

no agreement -- they must agree. This problem, interesting enough in its own right, sets
up the analytical agenda of the next chapter on efficiency and equity of the scored
template when some issues have more than two resolutions and where BATNAs and RVs
play an important role. Chapter 13 deals with Template Analysis for the special template
of the Fair Division problem and Chapter 14 generalizes this analysis to the general
template and where there are alternatives to negotiations present. In Chapter 13 we learn
that the source of joint gains stem mainly from the differences between the player's
evaluations of the template and this insight carries over into 14. The flavor of 14 is still
FOTE.

Chapter 15 deals with behavioral realities. We completely disengage from FOTE

and comment on laboratory behavior with student-subjects. What happens when subjects
without any training act as agents for principals who have provided them with a scored
template and precise reservation values stating the minimum acceptable contract values?
It took me by surprise when I first performed such simulations how very varied were the
final contracts. It's important to examine what seems to work and what doesn't.

Still in chapter 15, we next consider anomalies, biases and errors of behavior in
real-world negotiation settings. In an interactive setting misinterpretations beget
misinterpretations and a dynamic may ensue where the parties spiral downwards in their
pursuit of joint harms rather than joint gains. Cultural differences make it harder to
establish a constructive negotiation style even though cross-cultural differences in
interests are often the source for potential joint gains.

Chapter 16 takes a prescriptive/descriptive orientation. It offers partisan coaching

advice to an analytically inclined collaborator suggesting how he might negotiate with a
negotiator, who is less analytically inclined, who falls short of the FOTE ideal -- perhaps
falls in the POTE camp -- who claims prematurely and excessively. We suggest how to
negotiate in a hostile dispute with a disputatious adversary who needs to emote about
your client's past despicable behavior. Part of the strategy is to try to foster more
collaboration by effectively changing the others negotiation style.

PART IV: Interventions In Two-Party Disputes

Our first chapter in this part starts out by considering conventional facilitation and
mediation reserving the next chapter for conventional arbitration. Rather than getting
involved in disputes about nomenclature, we adopt the neutral language of “external
helper “ or simply “helper” or “3rd party intervenor”. In our abstraction, we talk about the
{A, B; H} dynamic where A and B are the negotiating parties and H the helper. We start
by drawing up lists of conventional roles that H might perform, saving those that have a
more analytical flavor for later development. . What role H plays depends on how H gets
involved (as invitee or invitor or a mixture of these), on the context of the problem, on
tradition, on the particular temperaments of the three actors involved, and on what is to
be negotiated.

Copyright  2003 Howard Raiffa 5

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Full Overview

We talk about the interests of H as another player -- but a different type of player.
We review the many reasons why external helpers are not used when perhaps they
should.

The chapter then considers more active roles for the mediator by involving H
more in generating proposals (e.g., in the form of Single Negotiation Texts) for the
consideration of A and B. We discuss President Carter's role at Camp David as a
proactive mediator with clout in the negotiations between Sadat of Egypt and Begin of
Israel. We also consider, as a second case study of proactive mediation, my intervention
in helping divide an art collection between two brothers who knew ahead of time that
they could not do it alone without jeopardizing their relationship.

We comment on the rapid growth of FMA (facilitation, mediation, and

arbitration) interventions and on the surge of interest in Alternate Dispute Resolution.
ADR constitutes a spectrum of interventions ranging from weak facilitation to strong
mediation and arbitration. In practice, a negotiated impasse often triggers mandatory
arbitration where H, in an evaluative mode, is expected to impose a "solution" that is
binding on the parties. There are all sorts of variations: arbitration may not be mandatory
but a voluntary option on the part of the negotiators; the proposal of H may not be
binding but suggestive; and in some circumstance H may be expected not only to be
neutral initially but to be non-evaluative throughout.

In the standard distributive bargain where A wants a higher value and B a lower
value, the conventional arbiter, after fact finding, proposes a final solution. In contrast, in
final-offer arbitration, A and B are required to submit sealed final offers and then the
arbiter, H, is required to select one of these offers. H has no other choice! We examine a
game-theoretic treatment of this type of final-offer arbitration and surprisingly find it
somewhat flawed from a normative perspective even though it seems to work in practice.

The latter part of the chapter deals with cases of complex integrative negotiations
in which H, acting as a neutral joint analyst (NJA), helps the parties achieve an ideal
collaborative compromise solution described at length in chapters 11 to 14. Such an NJA
acts as a special kind of non-evaluative arbiter. The possible reluctance of the parties to
truthfully reveal their reservation values (RVs) complicates the analysis and a double
auction bidding system is introduced to help resolve this complication.

After the completion of an unassisted negotiation, the parties realizing that there
may be still joint gains to be had, might choose to invite in an NJA to try to embellish
their agreement in a so-called post-settlement settlement. In this case the parties do not
have to reveal their RVs to the NJA -- the previously negotiated compromise acts as a
pseudo joint reservation value.

There remains the nagging question for the NJA: which point on the efficient
frontier should be selected as the most equitable. What is fair? The next chapter discusses
this question in some detail. In Chapter 13 and 14 we already introduced some candidate
rationales for "fairness" such as Nash's solution that maximizes the product of excesses

Copyright  2003 Howard Raiffa 6

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Full Overview

over RVs and the one that maximizes the minimum of the two proportions of potentials
(POPs).

The last chapter in this part examines the common nature of intractable disputes
(mainly between feuding contiguous countries or national entities or cultural groupings),
identifies a set of barriers that prevents constructive negotiations from taking place, and
suggests how parallel negotiations (track II) could help.

The chapter documents a successful intervention in the Peru/Ecuador conflict.

Influential, non-official surrogates from the belligerent countries were brought to
Cambridge, Massachusetts by the not-for-profit Conflict Management Group to engage in
a week-long activity called Facilitated Joint Brainstorming (FJB). The facilitator, Roger
Fisher, (backed by a team of helpers including myself) engaged the group in generating
creative options for the controversy, ruling out all claiming tactics. The surrogates were
evidently inspired enough at week's end to return to their countries and to become change
agents back at home. This is the way it was supposed to work and in interviews with the
subjects it seemed to have worked that way. Fisher was publicly commended by the
President of Ecuador and Foreign Minister of Peru for his contributions to peace in the
region.

This FJB exercise involved a logistical feat: getting 11 distinguished surrogates

(e.g., former foreign minister, or retired military chiefs of staff) around a table in a
neutral spot. As variant of the FJB which uses face-to-face discussion of surrogates, we
introduce Facilitated, Shuttle, Pre-Negotiations with Surrogates. The facilitator (perhaps a
dissertation writer) shuttles back and forth interviewing key surrogates on their
perceptions of the controversy, on their interests, on eliciting creative options, on
brainstorming in serial fashion, on obtaining their value tradeoffs, on determining their
BATNAs, and on identifying several -- not one! -- possible feasible contracts. The
facilitator can then prepare a pre-mediation briefing report (a PMBR) describing the
problem and hopefully some potential "solutions," -- i.e., some equitable and efficient
compromises The hope is that the PMBR may trigger a reconsideration of the merits of
trying to negotiate a solution to a festering dispute or motivate the consideration of an
opportunity that might be profitably negotiated.

PART V: MANY PARTIES (Chapters 21-27)

In this part we extend all fronts into many-party lands. By “many parties”, as in
some primitive tribes, we mean more than two.

Everyone knows that reaching a decision becomes harder the more people are
involved. Negotiations among multiple parties – or even decisions, which must be jointly
taken by a group which is ostensibly on the same side – are often long and unhappy
affairs.

Up to now we have been talking about relatively simple bilateral negotiations

(with the addition of a third party helper, who is not also a party at interest in the

Copyright  2003 Howard Raiffa 7

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Full Overview

negotiation). Once we move out of that setting, there are many different geometries we
could consider:
• A Group of Separate, Individual Negotiators
• Bilateral Negotiations with Multiple Participants on Each Side
• A Group of Advisors Preparing One Side for Negotiations
• A Permanent Decision-Making or Advisory Group
• An Ad Hoc Decision-Making or Advisory Group.

Each of these groups pose particular challenges. But there is also a core of similar
difficulties that will have to be managed in all of those contexts. Chapter 21 starts by
listing in what ways some groups do better than individuals and then why some groups
do worse. We examine some anomalies of group behavior. It then attempts to understand
the underlying reasons for this behavior – why do some groups behave so poorly. After
examining losses due to cognitive overload, poor coordination, poor communication, and
poor motivation, the chapter considers general prescriptive advice designed to partially
ameliorate some of these causes. The advice considers such matters as: membership in
the group; use of a facilitator or chair; need for an ongoing visual documentation of the
deliberations; role of brainstorming and devising; focus on purpose and choice of a
problem solving framework; decomposition of tasks and formation of sub-committees;
and the allocation of time.

The next chapter (22) considers groups that strive for consensual agreement. Each
of the members wishes to act in unison but there are strains in the community. Still they
don’t resort to voting or coalition formation. To a large extent this discussion generalizes
the two-party material in chapters 11-13 which has a strong FOTE flavor. As was the
case for two parties, we once again examine the fair division problem – but now with
more than two parties – before launching into our full analysis of feasible, efficient, and
equitable contracts for the many-party problem.

Chapter 23 has a game-theoretic flavor. It starts out by discussing the dynamics of

coalition building and I recount some of my experiences at IIASA where I often engaged
in that practice. We next consider the highly structured problem where every coalition is
valued at some specific, fully known, total monetary worth (with common knowledge of
this input data) and where players must then join coalitions and decide how to divide the
resulting joint revenue. It’s a wild game that defies prediction of outcome. Still some
modest advice can be given. We examine principles that can be employed (a favorite one
by Lloyd Shapley) to divide up the joint proceeds of the coalition of the whole in ways
that reflect the power of various sub-committees. We end our consideration of coalitions
by giving partisan advice to one party engaged in two separate intertwining negotiations;
each sets up a BATNA for the other.

Chapter 24 examines the use of voting procedures for group action. Such common
schemes as majority rules are fundamentally flawed since it can lead to intransitivities of
group preferences. We review Arrow’s famous Impossibility Theorem that proves that
there is no way for three or more (i.e., many) individuals to combine their individual,
ordinal preferences to obtain a group ordinal preference without violating some appealing

Copyright  2003 Howard Raiffa 8

Negotiation Analysis: The Science and Art of Collaborative Decision Making, Full Overview

desiderata. Many party negotiations often result in the identification of a few viable
contracts for adoption, and guess what? They vote. One way out of the dilemma is to
demand a richer set of inputs from the voters. They are asked not only about their ordinal
preferences but the intensities of their preferences. They are asked for cardinal orderings.
However, easier said than done – especially with the realities of insincere voting. We
introduce a case study that examines how a group of scientists selected a trajectory for
the Voyager mission to outer space.

The next chapter examines cases where at least one side in a negotiation is
non-monolithic. We imagine a two-party (external) negotiation across a table that is
complicated by the existence of an intense internal negotiation on each side of the table.
Contracts that are negotiated across the table may result in winners and losers on one side
of the table and the losers may try to block the negotiations unless they are duly
compensated by internal transfers. Some call it “bribes.” The challenge is to synchronize
internal and external negotiations.

In the next chapter we extend the material in Part IV on interventions in two-party

deals and disputes to many parties. We examine a hypothetical case of a community that
has to negotiate with a family that wants to use part of their extensive land-holdings to
erect a shopping mall and some high-rise office space. The quid pro quo is that the town
will get money from a higher tax base and some needed land that can be used in many
different ways. A facilitator/mediator is more active with the many-sided interests within
the town than with the division between the town and the developer. We conclude with a
case study of energy negotiations in the European Union. There were a series of
facilitator/mediators who vainly sought a feasible solution (for all the participating
countries), before the Luxenburgers stepped up to the plate and found the way toward
mutual acceptance.

We end our book on a philosophical note by going back to the two-person

prisoner’s dilemma game (or social trap) and introducing more players. The headings tell
the story: the commons problem (from sheep grazing to over-population of our planet);
on over-utilization of resources (like fishing and forests); on polluting the atmosphere and
despoiling the seas; on free riders (both individual, non-charity givers, as well as
countries that want everybody else to curtail their green-house emissions); on the Not-In-
My-Back-Yard (NIMBY) syndrome; on environmental injustice and on why
disadvantaged communities could exact enormous compensations for the acceptance of
an undesirable facility. And finally, all this is brought down to the personal level: “How
much are you willing to sacrifice for the good of the whole?”

Copyright  2003 Howard Raiffa 9

 Howard Raiffa: Short Vita (January, 2003)

Howard Raiffa is the Frank P. Ramsey Professor (Emeritus) of

Managerial Economics, a joint chair held by the Business School and the
Kennedy School of Government at Harvard University. He held professorial
positions in the Department of Economics and in the Department Statistics
and chaired the inter-university doctoral program in Decision Sciences. He
was the co- founder with Roger Fisher of the intra-university Program on
Negotiation located at the Harvard Law School.
As a scientific advisor to McGeorge Bundy and later to Philip
Handler, President of the U.S. National Academy of Sciences, Raiffa helped
in the negotiations ('67 - '72) leading to the creation of the International
Institute for Applied Systems Analysis (IIASA). The Institute, supported by
16 Academies of Sciences from the East (led by the Soviet Union) and the
West (led by the United States), was designed as a confidence-building
measure during the cold war. Raiffa was the first Director of the Institute
('72 - '75).
Raiffa received his Ph.D. in mathematics in 1951 from the University
of Michigan. He has worked in operations research, game theory, statistical
decision theory, decision analysis, risk analysis, behavioral decision theory,
and in conflict resolution and mediation and has received lifetime
achievement awards from each of those societies representing those fields.
He has supervised close to 100 doctoral dissertations. He holds honorary
degree from Harvard, Carnegie Mellon, Michigan, Ben Gurion, and
Northwestern Universities. In year 2000 he received the prestigious Dickson
Prize for Science (conferred annually to one recipient by the University
Professors of Carnegie Mellon University).
Supplement to Negotiation Analysis: The Science and Art of Collaborative Decision Making by Howard
Raiffa (with contributing authors David Metcalfe and John Richardson), Harvard University Press, 2003.

The Neutral Joint Analyst

Dear XXX and YYY;

I have been asked by ZZZ who just happens to know that you are
about to enter into very complicated negotiations. ZZZ thought I might help
you both and suggested I write to you and explain my craft. I bring to the
table skills of the intervenor (facilitator, mediator, and arbitrator), coupled
with the skills of a decision analyst. I call myself a Neutral Joint Analyst –
an NJA if you will – I would like to help you to find both an efficient and
equitable outcome of the negotiation opportunity you collectively confront.
By efficient I mean that you squeeze out the full potential of this negotiation
opportunity and not leave potential joint gains go unrealized sitting on the
table. By equitable I mean that I will treat you equally in seeking a
resolution of your joint decision problem – a resolution that gives each of
you a comparable fair gain over what each of you could gain acting
separately. We can think of this as aided collaborative decision making.

If you were to use my services – which because of my starving family

don’t come free – here is the way I propose to help you.

Stage 1. Formulating Your Joint Problem or Opportunity

I’ll help you jointly to formulate your collective problem and

opportunity. We jointly must decide just what has to be decided and decide
on the alternatives for resolution. In joint and individual meetings with the
two of you I will get to know the details of your problem and opportunity
and create a framework or template which specifies the issues to be
addressed and possible resolutions of each of these issues.
Along the way, from each of you, I might learn confidential material
that you would prefer not to disclose to the other. My reputation would be
destroyed if I were not to honor your confidential reflections most
responsibly. Absolutely no unauthorized leaks. Part of my comparative
advantage is that you might choose to tell me things that you would rather
not disclose to the other side so that in devising cooperative solutions of
your joint problem (or opportunity) I may be in the best position to devise
creative alternatives. I don’t want to come close to jeopardizing my
comparative advantage by telling them what’s for my ears only.

Copyright  2003 Howard Raiffa

The Neutral Joint Analyst

Stage 2. Evaluating What You Individually Want and Need

Working separately with each of you, I try to get you to express what
you really want and desire – your basic tradeoffs. Helping you to grapple
with tough questions like: how much of X do I need to justify giving up an
amount of delta on Y. Not easy, but, in a negotiation context, important
enough to have someone who can help you to think about horrendous
tradeoffs.
For efficiency and equity analysis to come, I’ll probe what might
happen if you can’ t come to an agreement with the other side, You’ll have
to think hard about what decision problem you might face if negotiations are
of no avail.
I’ll act under the assumption that what you tell me is the truth – no use
paying me if you tell me false info – but you may wish not to disclose some
particularly sensitive info. What I want from you is the truth but not
necessarily the whole truth. You’ll have to tell me just what I can disclose to
the other side. You are in control of the sharing of information. Most of my
clients are willing to share more info if they feel the other side is also
forthcoming.

If I’m to help you fully I would like to learn just what are your
possibilities if negotiations were to be aborted. I’ll encourage you, with my
help if you so desire, to explore your Best Alternative to No Agreement
(BATNA) to help you assess the minimum you should be willing to accept
in the actual negotiations.

Stage 3. Analyzing Your Joint Problem

I want to help you find a suitable compromise contract that is

perceived as fair and squeezes out the joint potential gains. We don’t want
you to agree on a contract when there is another one that you both prefer. If
you both are so inclined to tell me the whole truth, (confidentially) then I’ll
be in a position to suggest a compromise solution. From my perspective
finding such a “solution” is a mathematical programming problem. You
could jointly agree ex ante to bind yourselves to my suggested ‘solution” or I
could play the role of a post-settlement analyst: both of you come to a
compromise solution and then let me see if I can embellish it by finding an
alternate solution that you both might prefer. I can do all this not because
I’m smarter than you but I’m privy to information that each of you lack and
I’m a professional problem-solver.

Copyright  2003 Howard Raiffa 2

The Neutral Joint Analyst

To build up trust and confidence, why don’t you try me out on a easy
problem that is not critically important to you. If you want I could suggest a
problem that you both could practice on with me

. Your friendly neutral joint analyst,

Norman J. Altman

Copyright  2003 Howard Raiffa 3

Supplement to Negotiation Analysis; the Science and Art of Collaborative Decision Making by Howard
Raiffa (with contributing authors David Metcalfe and John Richardson), Harvard University Press, 2003.

Ch 10b Auctions And Bids B

Bidding Traps Arising from Ignoring Cues from the Actions of Others.
The case studies in this part of the chapter deal with the subtleties of
subjective probabilities. Player A might have an intitial probabilistic assessment of
some uncertainty, and after observing cues from the behavior of player B, A should
modify this distribution. Sometimes these cues are subtle and empirically ignored
and individuals are prone to fall into various auction or bidding traps. This will be
exemplified by three problems.
From an analytical perspective, the material of this chapter is related to the
discussion of Conditional Ambiguities and the Monty Hall conundra in Chapter 3.

PROBLEM 1: THE WINNER'S CURSE. COMMON UNCERTAIN VALUE WITH

DIFFERING PROBABILISTIC ASSESSMENTS
We’ll start this more involved analysis with special case. This time we’ll
assume that the bidders are involved in an auction where the prize has an uncertain
but commonly known value and that the bidders must grapple with privately held
probability distributions. In other words, the post-auction value of the prize to each
and every bidder is exactly the same. However, each bidder has a different
probabilistic assessment of what the post-auction value of the prize will be. They
keep this information secret and act on their beliefs. To set the stage for further
discussion we use two examples, one is frivolous and simple, the other real and
complex.
Two Examples
Example 1. A gallon-sized glass jar is full of pennies, nickels, dimes, and quarters.
The jar is transparent for all to see but no one knows the proportions of different
coins or the total number of coins. Each of 30 bidders, say, has his or her own
guesstimate of the value of the jar. The bidders can lift the jar if they wish. The
prize is the contents of the jar.

Copyright  2003 Howard Raiffa

Ch 10B Addendum On Auctions and Bids

Example 2. The prize is the oil deposits under a given site up for competitive bid.
All the bidders have some information about the deposits but different guesses
about the value. Let’s simplify a bit by saying with full information the value of the
oil would be the same for all bidders.
What’s the strategic essence in these two problems? First, this is the case of a
prize that has a common objective value, but--and it is a big but--the common value
is uncertain and each of many bidders have differing subjective perceptions of this
value.

The Coin-Packed Transparent Jar

Suppose that you have looked at the jar containing the coins and have
assessed your own subjective probability distribution of its true monetary value.
Your best guess is $200 but your distribution is quite spread out.1 You are bidding
against 29 other bidders. The number 29 is not critical, but the number should be
sizeable, not 3 or 4. Well, what do you bid? For the purpose of this exercise let’s
assume that you are risk neutral. Let me invite Jay and Ann to join this discussion.

Jay: It seems to me that if I’m risk -- or for the standard competitive bid -- I
neutral my RV should be $200. That’s as would shade down from my RV.
high as I would bid in the English auction, Ann: I agree so far with Jay. And since
and in the Vickrey auction I would bid my there are so many other bidders, I
RV of $200. Now for the English auction wouldn’t shade down too much from the
RV of $200 -- say $185.

1
Assume that $200 is the Expected Value of your assessed distribution.

Copyright  2003 Howard Raiffa 2

Ch 10B Addendum On Auctions and Bids

Jay: This seems like an easy case. Why Exp: Well you should suspect by now
single it out? that there will be a special message of
earth-shaking import.

Suppose the potential bidders recorded their single best guess at the uncertain but
common value of the prize. There would be 30 guesstimates and these would have
some distribution.
We still don’t know where the true value is. It could be below, or close to, or
above the average of these 30 best guesses. Suppose for now that the true value is
close to the average of these 30 best guesses. Now consider three of the bidders:
Alicia, Bernie, and Charlie. Alicia’s best guess is on the low end of the distribution
of best guesses. Bernie’s best guess is slightly above the average of the best guesses
and Charlie’s is extreme on the high side. Their subjective, assessed probability
perceptions of the true value are shown in Figure 10b.1.

Figure 10b.1 Three Subjective Probabilistic Perceptions of the True Value

Probability

A True B C Value

Copyright  2003 Howard Raiffa 3

Ch 10B Addendum On Auctions and Bids

The trouble is that Alicia, Bernie and Charlie do not know where their best guesses
lie among the distribution of best guesses. Now do you, Ann or Jay, see a problem
lurking here?

Jay: Well, if the bidders follow our more chance there is for a right tail outlier
advice and bid close to their RVs, then and if bidders bid close to their RVs then
Charlie will probably win the bid, or the winner of the uncertain prize will be
someone even more way out on the right paying too much.
tail will be the winner of the prize. Jay: The winner will be a loser.
Ann: Hey, this is a good deal for the Exp: That’s the message! If you win,
auctioneer. The more bidders we have the you lose.

THE WINNER’S CURSE

The strong desire to win a prize in the face of competition often
predominates over the rational analysis of what the prize is worth. Bidders ascribe
the act of winning a greater value than the assessed value of the winnings. For
example, a lot of oil companies that win highly contested bids, end up by paying too
much. This anomaly is encountered in innumerable auctions.

Exp: Well, what should you do about this guess knowing now that your best bid of
sad complication? $185 is the top of the 30 bids.
Jay: It’s not sad from the auctioneer’s Ann: Ah, that’s tricky. I would now
point of view. know that most of the others would
Exp: Agreed. I posed two questions to perceive the value of the common prize as
you for the coin-packed transparent jar below mine. If I thought that their
problem. Question A asks for your best opinions were somewhat informative, I
guess of the value of the prize before would now want to bias my best guess
knowing any of the other bids. And downwards.
Question B asks for your revised best

Copyright  2003 Howard Raiffa 4

Ch 10B Addendum On Auctions and Bids

Jay: I would want to bid an amount Ann: Sure, Jay, you said that you would
“X” say, so that if I were told my X value want to believe that you could expect a
were the highest bid, my revised best profit after knowing that your bid was
guess at the common value would still be highest. Can you analyze this problem
higher than X. That’s the way I would get further?
around the winner’s curse. Exp: Sure, but it’s the qualitative
Ann: This really is complicated. Let’s message that’s important. The game
assume that we originally had a theorists have a ball with this problem.
distribution centered around $200. That’s But in order for the game theorists to do
our best guess without worrying about their thing which is equilibrium analysis --
the strategic bidding problem. We then the problem has to be modified so that
said earlier that the more bidders there are each bidder gets confidential information
the closer we should bid to our RV. But and each is given an objective probability
now comes the winner’s curse. The larger distribution of the original RV’s of the
the number of other bidders the more others. It is too complicated for me to
worried we should be if we win. So with discuss in the elementary presentation.
29 other bidders we should be bidding
much lower than our $200 original RV.
We need a strategic RV like Jay just said.
Jay: Did I say that?

PROBLEM 2: AQUIRING A COMPANY; IGNORING INFORMATION

The “Acquiring a Company” exercise is the creation of Max Bazerman and
William Samuelson. Please read the caselet and after a few minutes of reflection,
answer the questions on the bottom of the page. When you’re ready, read on.

Copyright  2003 Howard Raiffa 5

Ch 10B Addendum On Auctions and Bids

ACQUIRING A COMPANY
In the following exercise you will represent company A (the Acquirer), which is currently
considering acquiring Company T (the Target) by means of a tender offer. You plan to tender in
cash for 100% of Company T’s shares but are unsure how high a price to offer. The main
complication is this: the value of Company T depends directly on the outcome of a major oil
exploration project it is currently undertaking. Indeed, the very viability of the Company T depends
on the exploration outcome. If the project fails, the company under current management will be
worth nothing - $0/share. But if the project succeed, the value of the company under current
management could be as high as $100/share. All share values between $0 and $100 are equally
likely.
By all estimates, the company will be worth considerably more in the hands of company A
than under current management. In fact, whatever the ultimate value under current management, the
company will be worth fifty per cent more under the management of Company A rather than
Company T. If the project fails, the company is worth $0/share under either management. If the
exploration project generates a $50/share value under current management, the value under company
A is $75/share. Similarly, a $100/share value under Company T implies a $150/share value under
Company A, and so on.
The board of directors of Company A has asked you to determine the price they should offer
for Company T’s shares. This offer must be made now, before the outcome of the drilling project is
known. From all indications, Company T would be happy to be acquired by Company A, provided it
is at a profitable price. Moreover, Company T wishes to avoid, at all cost, the potential of a takeover
bid by any other firm. You expect Company T to delay a decision on your bid until the results of the
project are in, then accept or reject your offer before the news of the drilling reaches the press.
Thus, you (Company A) will not know the results of the exploration project when submitting
your offer, but Company T will know the results when deciding whether or not to accept your offer.
In addition, Company T is expected to accept any offer by company A that is greater than the (per
share) value of the company under current management.
As the representative of Company A, you are deliberating over offers in the range of
$0/share (this is tantamount to making no offer at all) to $150/share. What offer per share would
you tender for Company T’s stock?
My tender price is: $_______ per share.

Copyright  2003 Howard Raiffa 6

Ch 10B Addendum On Auctions and Bids

******************************* (Pause)

Acquiring a Company: Simulation

Let’s thrash about a bit before we become systematic. How good would it be
to tender an offer of $60 per share? Well let’s see what happens. Since all values
between $0 and $100 are equally likely for Company T’s shares we ran 100
simulated trials based on values churned out by a random number generator. In this
experiment we varied the value of Company T to see what the expected return to
Company A will be when it offers T $60/share. The first twenty results of this
experiment are shown in Figure 10B.2. Column 2, under the heading, “Value to
Target” contains the random numbers between 0 and 100, column 3, “Value to
Acquirer” shows the value of T under the management of the Company A (e.g.,
18.29 x 1.5 = 27.42), column 4, “Acceptance” indicates whether T accepts A’s
offer, and column 5, “Payoff to Acquirer” logs the payoff to Company A from
acquiring (or not) Company T at the given price.

Copyright  2003 Howard Raiffa 7

Ch 10B Addendum On Auctions and Bids

Figure 10B.2 Acquiring a Company Simulation With $60 Offer

VALUE TO VALUE TO PAYOFF TO CUMULATIVE

TRIAL TARGET ACQUIRER ACCEPT ACQUIRER AVERAGE
1 18.29 27.44 TRUE -32.57 -32.57
2 71.89 107.84 FALSE 0.00 -16.28
3 53.62 80.43 TRUE 20.43 -4.05
4 53.31 79.97 TRUE 19.97 1.96
5 50.09 75.14 TRUE 15.14 4.59
6 30.7 46.05 TRUE -13.95 1.50
7 80.45 120.68 FALSE 0.00 1.29
8 79.66 119.49 FALSE 0.00 1.13
9 18.23 27.35 TRUE -32.66 -2.63
10 8.21 12.32 TRUE -47.69 -7.13
11 37.32 55.98 TRUE -4.02 -6.85
12 84 126.00 FALSE 0.00 -6.28
13 71.65 107.48 FALSE 0.00 -5.80
14 80.52 120.78 FALSE 0.00 -5.38
15 39.26 58.89 TRUE -1.11 -5.10
16 19.11 28.67 TRUE -31.34 -6.74
17 57.45 86.18 TRUE 26.18 -4.80
18 4.08 6.12 TRUE -53.88 -7.53
19 1.45 2.18 TRUE -57.83 -10.17
20 55.44 83.16 TRUE 23.16 -8.51

The first trial is a disappointing value to the target of $18.29 and the value to the
Acquirer is 3/2 x $18.29 or $27.42. A bid of $60 would yield an acceptance
(TRUE) and the net proceeds to the Acquirer would be $27.42 - $60.00 or -$32.58.
Not starting off so well. But on the third trial the Acquirer makes $20.43. Over 100
trials the average return is -$9.13. Not a good deal.
We also ran the same experiment but inserted an offer of $50 instead of $60.
We generated 100 new random numbers and the average return with this tender
offer is still disappointing: a dismal -$6.20 per share return. What’s cooking here?
The Acquirer should be able to make some money. Or should she?

Intuitive Thinking
Let’s examine how most people think about this problem. They start off
asking the natural question: What is the expected value per share of the company to
the Target? And the answer is $50. Next they ask: What is the expected value (per

Copyright  2003 Howard Raiffa 8

Ch 10B Addendum On Auctions and Bids

share) to the Acquirer? And their correct answer is 3/2 x $50 or $75. Next they say
if their RV is $75. what should they bid? They shade down a bit from $75 to $60
say.

Jay: Can I interrupt here? That’s just Jay: Well that will be $30 and not $50.
what I did. I thought I was clever. I bid That was my mistake. I didn’t condition
$51. Now what’s wrong with that logic? my probability assessment on Mr. T
Exp: Ann, what’s wrong? accepting my bid.
Ann: I did something similar. But when Exp: Now if Mr. A accepts your tender
you posed the question, “What’s wrong?” bid of $60, what is the expected value of
and after seeing the simulations, I think I the firm to you?
see what’s wrong. I should have guessed Ann: Well that would be 3/2 x $30 or
this from the beginning. I should have $45 per share. So a tender bid of $60
suspected. Why did you introduce this would result in an expected loss of $60 -
problem? It smacks of the winner’s curse. $45 or $15. That’s even worse than the
For really high values of the company to simulation indicated.
Mr. T., Ms. A is not going to sell the Exp: Your analysis is still not quite
company. Once Ms. A knows that Mr. T right Ann. The simulation is better than
accepts the tender offer her RV changes. you’re giving it credit for.
Jay: Oh, that’s clever. It is a disguised
version of the Winner’s Curse.
Exp: O.K. Let’s examine what happens
with a tender bid of $60. If Mr. T accepts
the bid of $60, what is the expected value
per share to Ms. T?

Copyright  2003 Howard Raiffa 9

Ch 10B Addendum On Auctions and Bids

Acquiring a Company: Decision Tree Analysis

It’s instructive to draw a decision tree of the problem from the Acquirer’s
perspective (see Figure 10B.3). At the first move, Bidder A must tender an offer
price per share. In the diagram we follow the branch corresponding to a $60 offer.
At the next move, Mr. T accepts or rejects the offer. As far as A is concerned this is
a chance move: T accepts if the value of the firm to A is less than or equal to $60;
and rejects the offer if x > $60. The chance of acceptance is the chance that x > 60
or 60/100 or 3/5.
If Mr. T accepts the offer of $60, then as far as A is concerned the next
move is at the control of chance and all values from $0 to $60 are equally likely. I
approximate this chance mode with 6 equally likely values of 5, 15, 25, 35, 45, and
55. The expected value to Ms. A if she offers $60 and T accepts is -$15. But now to
evaluate the offer of $60 we must move backwards in the tree to the node following
the branch where A chooses $60. The expected value at being at this node is
3 2
x(15)+ x(0),or − $9. Now that agrees pretty closely with the simulated result of
5 5

-$9.13.

Copyright  2003 Howard Raiffa 10

Ch 10B Addendum On Auctions and Bids

Figure 10B.3 Decision Tree Analysis for Acquiring a Company

X=5 7.5 - 60 = -52.5

-15
15 22.5 - 60 = -37.5
$0 ...
25 37.50 - 60 = -22.5
$30 Accepts
...
X =< $60 35 52.50 - 60 = -7.5
$60
A 45 67.50 - 60 = 7.5

$80 55 82.50 - 60 = 22.5

-9 Rejects $0
X > $60

Tender Price Per Share T's Response to Offer Value of X to Target Payoff to Acquirer

We can perform the same analysis for any bid b. Here’s the argument. Keep the
decision tree in mind. If b is greater than the value of the company, call that V, then
the offer will be accepted. But if b < V the company will reject your offer.

Whenever your bid is accepted your return is  V − b , where V is still uncertain.
3
2 
Since your bid was accepted any value in the range from 0 to b is possible and all
b
are equally likely. So on average the company is worth whenever your bid b is
2

3 b 
accepted, and we get    − b or  b − b  so already we see that this is a bad, bad
3
2  2    4 
deal. There is an analytical lesson here. Unaided intuition has limitations.
Simulating an actual play of the game can provide fundamental insights. Drawing a
symbolic decision tree can also help straighten out your thinking. Get used to these
techniques.

Acquiring A Company: Empirical Results

Bazerman and Samuelson asked graduate MBA for their tender offer bids per
share. The results are shown in Figure 10B.4. One hundred and twenty-three

Copyright  2003 Howard Raiffa 11

Ch 10B Addendum On Auctions and Bids

students responded when no money was involved. About three-quarters of them

offered prices over $50 per share for a negative return. The situation changed only a
bit when money was exchanged as demonstrated with the experiment with 66 other
students.
Figure 10B.4 Empirical Results for Acquiring a Company
NO INCENTIVES MONETARY INCENTIVES
BIDS (n = 123) (n = 66)
0 9% 8%
1-49 16% 29%
50-59 37% 26%
60-69 15% 13%
70-79 22% 20%
80+ 1% 4%

Below we plot the expected payoff as a function of the tender bid offer. It’s a deal
that Ms. A should shun.

Figure 10b.5 Expected Return To Acquirer as a Function of Bid

5
Expected Return to Acquirer

0
0 20 40 60 80 100
-5
($/share)

-10

-15

-20

-25

-30
Bid of Acquirer ($/share)

As Bazerman points out, not only do subjects do poorly on this exercise

but when the game is repeated over time they persist in their irrational
behaviors. That's because as shown in the simulations (Table 10B.2) some
returns are positive.

Copyright  2003 Howard Raiffa 12

Ch 10B Addendum On Auctions and Bids

PROBLEM 3: ASYMMETRIC INFORMATION

We’re now going to consider another special case and use some game
theoretic techniques to improve the value of our prescriptive advice. The structural
characteristics, or rules of the game, for this auction are that the prize is of uncertain
but common value (as it was with the transparent jar), that all bidders share the
same
objective probabilities about the ultimate value of the prize, but, and here’s the
twist, there are asymmetries in the information held by different bidders. At least
one of the bidders has “inside information” about the likely value of the prize.

Maxco-Gambit: Two Oil Companies Bidding for Drilling Rights

Let’s motivate this bidding case by bringing in two oil companies, Maxco
and Gambit, as our protagonists. Both oil companies have to submit sealed bids for
the rights to drill at a given site. Gambit, however, has privileged, confidential
information. Gambit owns the adjoining sites and has dug exploratory wells near the
boundaries of this site. So Gambit knows more than Maxco about the chances of
hitting oil and, if so, how much there might be. Gambit is careful not to share this
information with its competitor, Maxco.
We’ll abstract and simplify this situation into laboratory experiment. Imagine
that the prize is equally likely to be $0, $100, or $200. In the motivating oil problem
these are in units of millions or tens of millions of dollars. To be even more graphic
suppose that I, as the experimenter, have three sealed envelopes containing
respectively prizes of $0, $100, and $200. I choose one of these envelopes at
random; put it on the table in front of you and put aside the other two envelopes.
Now for the gimmick.
Assume that the bidders, labeled Maxco and Gambit, have differential
information. Gambit is shown the contents of the selected envelope. Gambit now
has perfect information about the value of the prize. Maxco knows that Gambit

Copyright  2003 Howard Raiffa 13

Ch 10B Addendum On Auctions and Bids

knows and Gambit knows that Maxco knows that Gambit knows and so on. Think
hard about this one.

What would you bid as Maxco? And what would you bid as Gambit conditioned on
what you might learn about the contents of the envelope?

********************(Pause)
Let’s play the conventional sealed-bidding game: high bidder wins at the
high price. When you are ready please read on. O.K. Let’s resume. Let me call in
Jay and Ann again.

Exp: What did you bid as Maxco? were $100 but you would lose $63 if the
Jay: As Maxco I bid $63. prize were zero.
Exp: And Ann, what do you do as Jay: You’re right. But you were not so
Gambit? smart, either. You would bid $112 for the
Ann: My strategy was to bid $0 for the $200-prize. You could have squeezed
$0 prize; $51 for the $100 prize, and harder and come in with some value like
$112 for the $200 prize. $65 say.
Exp: So in this case Jay playing Maxco Exp: Well, what do we see here? If
would win the prize if it were $100 (since Gambit is very conservative -- i.e., bids
his bid of $63 is higher than her bid of close to $100 for a $100 prize and close
$51), but he would lose the $200 prize to $200 for a $200 prize -- then Maxco
(since his bid of $63 is less than her bid of will only win when the actual prize is
$112). zero.
Jay: Good. I would capture the $100 Jay: This is really the winner's curse
prize. once again. Maxco wins the prize only to
Ann: Why are you so pleased with lose money. So the conclusion seems to
yourself? You would win $37 if the prize be that Maxco shouldn’t bid. Especially
against a conservative Gambit.

Copyright  2003 Howard Raiffa 14

Ch 10B Addendum On Auctions and Bids

Ann: But if Maxco thinks this way, then Ann: And how should I behave if I
I, as Gambit, shouldn’t be conservative. I don’t have this empirical information
should bid $30 or a $100 prize and say about others?
$40 for a $200 prize. Exp: That’s the usual case, of course.
Jay: But be careful Ann. If I suspect You just have to imagine yourself in the
that you will be greedy and bid so low, I, role of the other party; try to think as you
as Maxco, will be able to sneak in with a believe he or she might think; and
bid of $50 and get both the $100 and heroically assess your probabilistic betting
$200 prize. distribution of what the other party will
Jay: It is important for us as a player, do. I like to think of l00 different people
to try to out-psych the other player. But each playing the role of the other party
where does this end? and I assess the distribution of what I
Exp: Well, one tack would be to think they will do.
examine equilibrium behavior. The idea is Jay: But if you’re lousy at this guessing
to give advice to both players that will be game, you won’t do well will you?
stable. The advice should be such that Exp: That’s right. But I think you are
each player would want to follow the better off thinking this way than not. Of
advice if the other player follows the course, some laboratory experimental
advice. The other tack is to examine results bolster my impressions.
empirically how people like yourselves
behave so that you could behave best
against this empirical mix.

Before continuing I want to recount an experience I had in a large MBA class

in Negotiation Analysis. Early in the semester I described the Maxco-Gambit game
and chose two students at random to play the game for money in the class. I was the
banker. I’ll use fictitious names: Max for Maxco, and Gambriel for Gambit.
I shuffled three envelopes with prizes $0, $10, and $20, chose one at random,
and showed it to Gambriel. Then Max and Gambriel wrote down their bids. “Well,
what did you bid Max?” I asked. “Fifteen dollars,” replied Max and there were

Copyright  2003 Howard Raiffa 15

Ch 10B Addendum On Auctions and Bids

impolite guffaws from the 100 onlookers. “And you Gambriel?” I continued. “I bid
$12,” said Gambriel, “because the envelope indicated the $20 prize. It turns out to
have been a stupid bid.”
Max was delighted. He netted $5 and snickered back at his disapproving
audience. This shows the danger of a sample of size one. Here is a case of a very
bad decision for Max that turned out to have a good outcome. Some students took
Max’s side and said that if the outcome was good, the decision was good. Max said,
“I’m lucky in gambles; I had an intuitive flash that the chosen card was $20. So I
gambled and won.” I thought to myself, “I wouldn’t hire this guy!”.

Think carefully about the Maxco-Gambit game described above, then answer the
following five questions:

1. If you were Gambit, would you prefer to obtain your confidential information
secretively (so that Maxco would not know you had this privileged information) or
openly (so that Maxco knew that you knew the value of the prize and he or she
didn’t)?
2. If you were Maxco and by counter-intelligence, say, you could also learn the value
of the prize, would you want to learn this information secretively (so that Gambit
would not know you also knew the value of the prize) or openly (so that he or she
would know you knew)?
3. If you were Gambit and you were not given the confidential information about the
value of the prize but could buy this information (in a way that Maxco would know
you knew), up to how much would you pay?
4. Neither Maxco nor Gambit knows the value of the prize. But this information is
available on a confidential basis to the highest bidder. What would be your RV for
this auction?
5. There is a sequence of repeated Maxco-Gambit games. At each trial the high
bidder at that trial gets the privileged confidential information at the next trial.

Copyright  2003 Howard Raiffa 16

Ch 10B Addendum On Auctions and Bids

This means, for example, that if Maxco outbids Gambit on trial 4, then Maxco
learns the value of the prize at trial 5 and Gambit now knows at trial 5 that Maxco
knows the value of the prize and she doesn’t. How would you bid?
***********************

Ann: I, Gambit, would want Maxco to that problem I would assess a probability
know I have the information. I would distribution for the bid of Maxco. Then I
want to intimidate him to bid zero so that could figure out my best retort and
I could bid low. Indeed, even if I could calculate my expected return. Next I
not get the information, I would want to would consider the case where I learn the
lead him to believe I had the information. value of the prize and I know Maxco
Exp: Good. Now for question 2. knows that I know. Once again I would
Jay: I, as Maxco, would want my assess a probability distribution for
counter-intelligence to be secret. I would Maxco bids and then calculate my best
like her to believe I was in the dark, so $100 and $200 responses and finally
she would be greedy and bid low. Then, if calculated the value of that game. The
I learned the prize were $200, and if she difference would inform me about the
didn’t know I knew, then I could steal the value of this information. Go on to
$200 prize for $70 or $80. question 4.
Exp: I agree. Now for question 3. Jay: My turn. It’s a little different from
Ann: I’m Gambit so I suppose you want question 3 because if I don’t win the bid,
me to respond. But I don’t have the she will win it and clobber me. I have to
faintest idea what the answer is. Is the figure out my RV for this information
answer obvious? with the understanding that if I have it I
Exp: No, it’s not. But the line of attack can intimidate her while if she has it she
should be clear. Here’s what I would do can intimidate me.
to answer the problem. I would first Exp: That’s right. Go on to question 5.
consider the problem when I, Gambit, did Ann: I would want to bid higher than
not know the value of the prize and usual because if I win, I could intimidate
Maxco knew I didn’t know. To answer Maxco at the next round. It would

Copyright  2003 Howard Raiffa 17

Ch 10B Addendum On Auctions and Bids

depend on the number of rounds to be Exp: All right, if you want to do this,
played. This would be fun to play. I’ll be the bank, but let’s play it for
pennies instead of dollars.

Asymmetric Imperfect Information

These Maxco-Gambit bidding exercises were motivated by the real-world
problem of bidding for oil: the right to drill at a given site. As you recall Gambit had
privileged information, but not necessarily perfect information. In the exercises we
just covered, we assumed the case of the polar extreme where Gambit learns the
exact value of the prize. We now back away from that extreme assumption and to
capture this complexity we assume that Gambit doesn’t learn the contents of the
envelope that determines the prize, but learns something less precise. We now
reduce the size of the prizes tenfold and use three envelopes with prizes $0, $10, or
$20; and one is taken at random to determine the actual prize. Now assume that
Gambit learns the contents of one of the two discarded envelopes. And assume, as
before, that Maxco sees that Gambit has this additional information. For example, if
Gambit learns that one of the discarded envelopes is $20, then she knows that the
prize is equally likely to be $0 or $10. The information is valuable, but how
valuable? How intimidated should Maxco be in this case? What’s the expected
value of the game now to Maxco?
Some of you, intrepid souls, might want to try an analysis yourselves. Others,
and I guess the majority of you, might want to take a respite from these Maxco-
Gambit games and push on with the fascinating material still to come.

Maxco-Gambit: Illustrative Bids

Let’s explore some possible bids. See Figure 10B.6. Let m represent
Maxco’s bid and G0, G10 and G20 represent Gambit’s bids if she learns the contents
of the prize are $0, $10, and $20 respectively. Case 1 examines the results when m

Copyright  2003 Howard Raiffa 18

Ch 10B Addendum On Auctions and Bids

= 10, G0 = 0, G10 = 8, and G20 = 11. What happens to Maxco in this case? Well
Maxco would lose his bid of $10 if the prize is 0; wins $0 if the prize is $10; and
he does not win the bid if the prize is $20 -- that is because Gambit’s bid G20 is 11
which is higher than m = 10. So Maxco’s expected winnings is:
1 1 1
x(−10)+ (0)+ x(0) or
3 3 3

-10/3. Now glance down the seven illustrative cases and reflect a bit on what you
see happening.

Figure 10b.6 Illustrative Bids for Maxco-Gambit

CASES m G0 G10 G20 EV(m)

1 M
10 0 8 11 -M
-10/3
2 10 0 7 9 0
3 5 0 7 8 -5/3
4 5 0 3 7
5 5 0 3 4 15/3
6 5 0 3 4 -2/3
7 5 0 1 1.5 24/3

From this simple analysis we can already see that if Gambit is conservative (i.e., G10
near $10, and G20 near $20) Maxco loses money when he or she has bid. By
contrast, if Gambit squeezes hard (i.e., G10 and G20 are very low), then Maxco can
make some money by bidding 5 or so.

Maxco-Gambit Decision Trees

We now proceed to a decision tree analysis of the bidding problem with
asymmetric information. First we examine Maxco’s decision tree (see Figure
10B.7).

Copyright  2003 Howard Raiffa 19

Ch 10B Addendum On Auctions and Bids

Figure 10B.7 Maxco's Decision Tree

Maxco's Payoff

0 C
-m

1/3 M wins
B 10 - m
G10 < m
A m 10 D
1/3
M loses
0

1/3
M wins
20 - m
20 G20<m
E

M loses
0

Let’s consider Maxco’s mathematical machinations. Maxco must submit a

bid m. Then, at chance node B, there is a one-third probability of yielding a prize of
$0, $10, or $20. Let’s examine node D. We imagine that Gambit knows the prize is
$10 and submits a bid of G10. If that bid, G10 is less than m, Maxco wins l0-m; if
G10 > m, then Maxco loses and gets a zero pay off. The expected payoff at node D
is therefore
(10-m) x Prob [Gl0 < m] .
Arguing in a similar fashion, the expected payoff at node E is
(20-m) x Prob [G20< m] .
The expected payoff for Maxco is the average of the payoffs at nodes C, D,
and E. For various m-values, Maxco must think of the possibilities that G10 and G20
are below m.

Copyright  2003 Howard Raiffa 20

Ch 10B Addendum On Auctions and Bids

Now let’s look at Gambit’s decision tree (see Figure 10b.8). At node A,
chance yields prizes of $0, $10, and $20. At node B, Gambit, knowing the prize is
worth $10, must choose a bid G10. At node D, Gambit wins if Maxco’s bid m is less
than G10. A similar story for nodes D and E. Gambit must assess probabilities that
Maxco’s bid is less than G10 and G20.

Figure 10B.8 Gambit's Decision Tree

G's Payoffs
0
0
G wins
1/3
10- G10
B D G10 > m
G10
A 10
1/3
G loses
0

1/3 G wins
20 -G20
C
G20 > m
20
E
G20 E

G loses 0

Copyright  2003 Howard Raiffa 21

Ch 10B Addendum On Auctions and Bids

Maxco's uncertainty involves the unknown values of G10 and G20. Should
Maxco work through Gambit’s tree to figure out what Gambit will do? This is not
easy because Gambit’s tree depends on the unknown Maxco bid m. How far should
we go thinking about what each of us is thinking? Game theorists want to go the
whole way. To think what the other is thinking about what you’re thinking about ...
etcetera. In practice most people have trouble going even the first step in new
complicated problems. Of course, if the problem is very clear cut, important and
repetitive, it may be necessary to do more interactive, reflection analysis and then
equilibrium analysis becomes of real practical importance. Let’s see what happens
empirically with the Maxco-Gambit game in the laboratory.

Empirical Distribution of Maxco and Gambit Bids

We can learn something by studying the way people bid in practice - even if
its not a real bidding situation. Here are the results from a class of eighty-six
Harvard MBAs. The average Maxco bid was $4.82 with a standard deviation of
$0.84. Note that 25 students bid $0 and 4 bid over $10. The center graph depicts the
distribution of Gambit bids after Gambit learns the prize is worth $10. The modal
bid is $9; the average bid was $6.73 with a standard deviation of $1.04. One
confused soul bid $11. The bottom graph depicts the distribution of bids of Gambit
after Gambit learns the prize is $20. In this case $10 was the modal bid and 34
students bid higher than $10. The average bid was $10.14 with a standard deviation
of $2.12.

Copyright  2003 Howard Raiffa 22

Ch 10B Addendum On Auctions and Bids

Figure 10B.9 Empirical Results of Maxco-Gambit Bids (Harvard MBAs)

Maxco Bids with Unknown Prize

30

25

No. of bids in range 20

15

10

0
0 1 2 3 4 5 6 7 8 9 10 11 12
Maxco Bid ($)

Average bid $4.82 Standard deviation $0.84

________________________________________________________

Gambit Bids with a $10 Prize

20

16
No. of bids in range

12

0
0 1 2 3 4 5 6 7 8 9 10 11
Gambit Bid ($)

Average bid $6.73 Standard deviation $1.04

________________________________________________________

Copyright  2003 Howard Raiffa 23

Ch 10B Addendum On Auctions and Bids

Gambit Bids with a $20 Prize

32
28

No. of bids in range

24
20
16
12
8
4
0
0 2 4 6 8 10 12 14 16 18 20
Gambit Bid ($)

Average bid $10.14 Standard deviation $2.12

___________________________________________________________

Expected Payoff To Maxco

Figure 10b.10 plots the expected return to Maxco against the empirical
distribution of Gambit bids. The curve is a bit irregular but the message is clear.
Positive Maxco bids yield negative expected payoffs! Against the empirical mix of
Gambit bids, Maxco should bid $0.

Copyright  2003 Howard Raiffa 24

Ch 10B Addendum On Auctions and Bids

Figure 10B.10 Expected Payoff to Maxco

0
0 2 4 6 8 10 12

Expected Payoff to Maxco

-0.5

-1

-1.5

-2

-2.5

-3

Maxco Bid ($)

Expected Return to Gambit

Figure 10b.11 plots the expected return to Gambit when Gambit learns that
the prize is worth $10 of Maxco bids. It is computed against the empirical mix of
Maxco bids. The expected return to Gambit is rather flat for bids of $6 or less -- it
peaks about $5.50. The $0 bid for Gambit is really more like $0.10; it beats all
strictly zero bids of Maxco. Bids above $6 rapidly deteriorate. The actual modal bid
of $9 for Gambit yields an expected return (against these Maxco bids) of $0.75.

Copyright  2003 Howard Raiffa 25

Ch 10B Addendum On Auctions and Bids

Figure 10B.11 Expected Payoff to Gambit With a $10 Prize

Expected Payoff to Gambit ($)

2.5
2
1.5
1

0.5
0

-0.5 0 2 4 6 8 10

-1
-1.5

Gambit Bid ($)

Expected Return to Gambit When the Prize is $20

Figure 10b.12 plots the expected return to Gambit when Gambit learns that
the prize is worth $20. It is computed against the empirical mix of Maxco bids. The
best bid for Gambit is a little over $10 -- just enough to outbid a lot of those
misguided Maxco bids of $9 and $10. Very conservative bids of Gambit above $11
rapidly lose expected payoff for Gambit.

Figure 10B.12 Expected Payoff to Gambit with a $20 Prize

10
Expected Payoff to Gambit ($)

9
8
7
6
5
4
3
2
1
0
0 2 4 6 8 10 12 14 16 18 20

Gambit Bid ($)

Copyright  2003 Howard Raiffa 26

Ch 10B Addendum On Auctions and Bids

Now let’s reflect about what all this means. Let me start a dialogue with Ann and
Jay.

EXP: Well, what do you think? What do What do you think happens after I show
these last three pages, depicting the these results to the bidders and then they
empirical results, tell us? play the game a second time? Why don’t
Ann: Some of your students were you think about this before reading on.
confused.
EXP: Granted. You always have to
expect some modicum of confusion and
misunderstanding. But that happens in the
real world as well as in the laboratory.

After Feedback
EXP: I had my students replay this game would be in equilibrium? The advice to
and what do you expect happened? each must be such that neither bidder
Ann: A lot more Maxco players would would want to change knowing the other
bid $0. would follow our advice.
EXP: That’s absolutely right. And Ann: I don’t see how you could do this.
Gambit players? If you tell Maxco to bid any amount like
Jay: I would bet they would bid a lot $3.00, then Gambit would want to let gl0
lower than they did before, especially for and G20 equal to $3.10. But if Gambit
the $20 prize. were told to bid $3.10, then Maxco
EXP: That’s also true. And it also turns would want to bid $3.20.
out that some Maxco players who live Jay: It seems to me any advice to
dangerously and bid about $5.50 steal a Maxco would be unstable if Gambit knew
few big prizes from very aggressive it.
Gambits. EXP: So far I agree with you, but you
Let’s speculate. What kind of operational haven’t thought hard enough yet. The
advice can we give to the players that trick is to tell Maxco and Gambit to use

Copyright  2003 Howard Raiffa 27

Ch 10B Addendum On Auctions and Bids

randomized procedures. It can be shown, EXP: Yup, you’re right. But if Maxco
but I won’t do it here, that it is possible to were advised to bid zero, then Gambit
give randomized advice to each that yields could bid nice and low and we would go
an equilibrium situation. around in circles once again. The only
Jay: I’m not sure I follow you. You equilibrium situation is as I indicated.
would tell Maxco to base his bid on a Look, we’ve gone into this equilibrium
random drawing from some distribution. behavior more than I think it warrants, so
Like all bids should be equally likely let’s push on.
between $0 and $10. Is that it? Jay: I’m not satisfied
EXP: Or to bid $0 with probability 0.4 EXP: Let’s consider the use of shut-out
say, and the spread the other 0.6 bids. Would it ever be advantageous for
probability uniformly over the interval of Gambit to announce her strategy to
$0 to $10. That’s the idea. Jay. Maxco before the game starts? By
Ann: And you’re saying that if Gambit announcing her strategy, she would
learns about this advice, but doesn’t learn declare exactly what she would bid if the
about the particular random drawing, then prize V equals $0, equals $10, equals $20.
she should also follow your advice. But Ann: I can’t see how this could help
your advice to her must also involve Gambit.
random elements or else Marco could EXP: Suppose Gambit announces that
sneak in with a non-randomized response. she will bid $0 if V=0; will bid a bit over
EXP: That’s the idea. Such equilibrium $5.00, say $5.10, if V=$l0; and will bid a
advice is shown in an appendix.2 It’s not bit over $l0, say $10.10, if V=$20.
so easy to derive. Jay: Well, if Maxco believed Gambit,
Jay: You indicate on the bottom of the then Maxco could bid just higher than
page that if Maxco follows your advice, $5.10, say $5.20.
then his expected return is zero. That’s EXP: What would Maxco’s EV be in
scary. Why bother? Why not just bid that case?
zero? Jay: Well, if V were $0, he’d lose
$5.20; if V were $10, he would win

2
$4.80; if V were $20, he would lose the
Not included in the preliminary version.

Copyright  2003 Howard Raiffa 28

Ch 10B Addendum On Auctions and Bids

auction and get zero. Since each of these his shut-out bid, then Maxco would steal
is equally likely, I guess his EV would be the bid.
negative. Ann: But why would he do that? He
Ann: Maxco would be best off bidding would lose $10.20, if V were 0; lose 20¢,
zero. if V were $10; and gain $9.80, if V were
EXP: Gambit’s publicly announced $20; which would yield a negative EMV.
strategy is called a shut-out bid. It is EXP: How negative?
meant to intimidate Maxco. Ann: Not much. About negative 20¢.
Ann: But if the intimidation were to EXP: And what happens to Gambit?
work, Gambit could surreptitiously bid Jay: He would get nothing. Ah, I see
just above $0.00, say 10¢ if V were $10 what’s going on. If Gambit can publicly
or $20. announce a shut out bid Maxco can really
Jay: Ah yes, but Maxco might expect sock it to Gambit and lose only 20¢ in the
that Gambit is unprincipled and conniving deal. And perhaps Maxco can demand
and bid say 40¢. some of the payoff to behave nicely.
EXP: If Gambit threatened you, playing EXP: All I’m trying to do is to stir up
the role of Maxco, with intimidating shut- the pot a bit. Shut-out bids should not be
out bid, and if you could have a pre-play taken lying down by the Maxcos of the
discussion with Gambit what would you world allowing all the gain to go to the
say to him? intimidating party. Finally you might want
Ann: I don’t see what Maxco could do. to recall the discussion of the Ultimatum
EXP: How about if Maxco threatens to Game of Chapter 4.
bid $10.20? If Gambit were locked in to

Copyright  2003 Howard Raiffa 29

Ch 10B Addendum On Auctions and Bids

An exercise for the mathematically inclined: Consider the Maxco-Gambit Problem with just two
equally likely payoffs: $0 and $100. As before Gambit has privileged information and knows the
value of the prize. There’s common knowledge. Find the pair of randomized equilibria strategies.
Hint: Try the randomized strategy for Maxco that puts some weight on a bid of $0 and distributes
the rest in a way that makes any pure strategy of Gambit optimal in the range of $0 to $50. Try a
similar trick for Gambit. The value of the game for Maxco turns out to be a disappointing $0. But
nevertheless bidding $0 for Maxco cannot be a part of an equilibrium pair.

Core Concepts
We consider the distributive negotiation problem when a single seller is confronted with
several potential buyers. (Or a single buyer with several sellers.) Face-to-face negotiations can now
be replaced by several types of auctions or competitive bidding mechanisms. We introduce several of
these designs and examine them from both a game-theoretic, decision analytic, and behavioral
perspective and both from the orientation of a bidder and the auctioneer.
In the open, ascending outcry auction a bidder should (but often doesn't) prepare for the
auction by determining her breakeven value. There is no need for an analytically inclined bidder to
assess a probability distribution of the Maximum Bid Of Others -- a so-called MBOO analysis for
this case. A MBOO analysis is central to the decision analytic approach - but not game theoretic
approach -- of the Dutch (descending) auction and the competitive, sealed-bid mechanism.
The high-bidder-wins-at-the-second-high-price auction (or philatelist or Vickery auction) has
the nice feature that it is optimum to bid one's breakeven value regardless of what other bidders
choose to do. It does not require a MBOO analysis to act optimally.
The chapter then considers two special cases: (1) reciprocal buy/sell bids that are especially
useful when two business partners seek to dissolve their partnership; (2) combinatorial bids such as
the FCC auction when the federal government auctioned off 2500 licenses for radio frequencies. The
complicating feature is that for many bidders the value of a given lease depends on the other leases
that bidder wins.

The central theme of Part B deals with subtle issue of conditional probabilities. There is
often information to be gleaned from the prior choices of others.

Problem 1 considers the case where the prize is common but uncertain. In bidding for a transparent
jar filled with coins, the winner of the bid would very often be someone who had an extreme
perception of the value of the jar. The winner often finds out that he or she has paid too much. Thus
the winner (of the prize) turns out to be a financial loser -- thus the "Winner's Curse."

Problem 2 considers the (Bazerman-Samuelson) case where an Acquiring firm submits a bid to a
Target firm that can either accept the offer or reject it. The target firm has privileged information
about the value of the firm. Thus, the acquirer's conditional probability distribution for the value of
the firm should depend on whether the target accepts or rejects the offer. In the case discussed, the
acquirer should not want any deal the target would accept. This subtlety is missed by most subjects.

Copyright  2003 Howard Raiffa 30

Ch 10B Addendum On Auctions and Bids

It's another case of subjects becoming confused about conditional probabilities and is somewhat
related to the Monty Hall example in Chapter 3.

Problem 3 is abstracted from the case of two oil wildcatters bidding for a lease when one bidder has
privileged information about the chances and extent of the oil deposits.
In the Maxco Gambit Cases both bidders have identica1 probability distributions of the value of the
prize but before the bidding takes place Gambit gets privileged information about the value of the
prize. This raises the question of the value of information in a game setting. Part of the value comes
from an intimidation factor: the less knowledgeable player may be wise not to bid or to bid low and
the party with the added information can exploit this knowledge.

One Appendix does an equilibrium Analysis of the both-pay competitive sealed-bid problem.
The both-pay ascending auction, or the so-called escalation game, was featured in Chapter 9.
Another Appendix does equilibrium analysis for a slight simplification of the Maxco-Gambit
Problem. In this case it is equally likely that the prize is $0 or $100 and Gambit, not Maxco, has this
privileged information.

Appendices

A. On Choosing a Random Quantity with a Given Cumulative

Distribution Function

Let X denote the random quantity (variable) with CDF F(.). By definition, for
any constant x,

F(x) = Probability { X ≤ x } .
In order to select a random value for X governed by F, we proceed as
follows:
Let R be a random number on the interval from 0 to 1.0 – all values on
the unit interval being equally likely. Independent R-values can be generated
in EXCEL by the command +RAND. For any single R drawn,
say r, obtain corresponding x from the equation
r = F(x) = Probability { X ≤ x } ..
Let

Copyright  2003 Howard Raiffa 31

Ch 10B Addendum On Auctions and Bids

2
F(x) = x , for 0 ≤ x ≤ 1.

To find a random number, x, governed by F, we solve

2
r= x

or
½
x=r ,

where r is a random number on the interval from 0 to 1.

B. The All-Pay Competitive Bid with a Common Known Prize

Rules of the Game. Each of n + 1 players must submit a sealed bid for a $100
prize. The top bidder wins the $100 but all bidders sacrifice the value of their
bid.
Illustration: Let n =2 so that there are three bidders. Assume the bids are as
shown :

Bidder Bid Payoff

A $23 - $23
B $48 +$52
C $19 - $19
What would you bid ?
Equilibrium Analysis.
A little reflection indicates that the players must choose a randomized
(mixed) bid on the interval from $0 to $100. We seek a symmetric solution

Copyright  2003 Howard Raiffa 32

Ch 10B Addendum On Auctions and Bids

where each bidder uses a common Cumulative Distribution Function F. If the

other n players each choose F and you bid an amount b, your expected value
payoff is
EV(b | F ) = - b + 100 Prob{ all n bidders choose values ≤ b }
n n
= - b + 100 [F (b)] = - b + 100 F (b).

Now any bid b in the interval from $0 to $100 must be a good bid against this

common F, or else you yourself would not use F; hence it must be that

EV(b | F ) = constant, K (say),

so that
n
- b + 100 F (b) = K , for all 0 ≤ b ≤ 100 .

Now since, by definition, F(100) = 1, we must have

- 100 + 100(1) = K , or K = 0,
and substituting this value for K we get
n
F (b) = b/100

or
1/n
F (b) = [ b/100 ] .
To use the mixed (randomized) strategy F, you would generate the
random number r and solve
1/n n
r = [ b/100 ] or b = 100 r .

So we see that as the number of bidders increase, you should bid more
cautiously but your equilibrium expected payoff remains 0 for all n.
For n =1 , i.e., you are bidding against one other bidder, you should
choose equally likely bids from $0 to $100. If you choose bid b, you get a b/100
chance of winning 100 so that your expected up side just cancels the downside.

Copyright  2003 Howard Raiffa 33

Ch 10B Addendum On Auctions and Bids

This does not look like a good game to play. So many bidders will
n
either not bid (i.e., choose b = 0), or choose b < 100 r , and in this case

perhaps you should bid even more aggressively than the equilibrium theory
suggests. But if this logic appeals to you, it might appeal to others and
therefor perhaps you shouldn’t bid after all. But we can go on … but we won’t.

Copyright  2003 Howard Raiffa 34

Supplement to Negotiation Analysis; the Science and Art of Collaborative Decision Making by Howard
Raiffa (with contributing authors David Metcalfe and John Richardson), Harvard University Press, 2003.

Ch 14 Addendum On Risk Sharing

Mr. George is an oil wildcatter. His past diligence and business

acumen have assured him a good reputation, and he now enjoys the right

to drill for oil at a given site. The trouble is that he has liquidity problems:

most of his money is tied up in other risky ventures and his credit rating at

the bank is not favorable. The cost of drilling is uncertain, but he has the

possibility of taking seismic soundings at the site which will yield some

information—but not perfect information—about the possibilities of

finding oil. He could plunge all his financial resources into this deal and go

it alone; or he could borrow more money at the bank; or he could cut

others into the deal, either by means of a straight proportional sharing of

profits and losses or, perhaps, a different proportional sharing on the up

and down sides of the deal. Let's assume for the moment that all

contingent, financial sharing arrangements are secure—that all contracts

are inviolable both in law and in the intent of the protagonists—and look

at one way in which this problem can be abstracted into a risk-sharing

negotiation problem. The terrain we're about to enter into is so vast,

including as it does financial markets, equity financing, insurance, and

reinsurance, that we must be careful not to get lost in its intricate byways.

Copyright  2003 Howard Raiffa

Ch 14 Addendum On Risk Sharing

Mr. George approaches Mr. Lloyd, a speculator, to share his risky

venture with him. They examine their options and identify one strategy

that appears promising, but the payoffs are uncertain: these depend on the

(uncertain) cost of drilling, on how much oil is down there, on how easy

the oil is to recover, on future regulations, on future oil prices, and on a lot

more. To simplify, we'll say that they depend on which one of five states of

the world—A, B, C, D, or E—will prevail. (If we were to be more realistic,

we might use something like five thousand states of the world.) Mr. Lloyd

consults his own experts and obtains probabilistic assessments of the five

potential outcomes; these differ from Mr. George's assessments as shown

in Table 14A.1. All assessments are kept confidential.

TABLE 14A.1. Potential outcomes and sharing rules.

Copyright  2003 Howard Raiffa 2

Ch 14 Addendum On Risk Sharing

George and Lloyd, however, do agree on the financial implications

of the deal, conditional on a given future state prevailing.1 A dry hole

(states), for instance, would lead to a loss of $70,000, or an abortive attempt

after a negative seismic sounding (state B) would lead to a loss of $20,000.

George and Lloyd have to agree on how to share the financial

proceeds in each of the five states. If state A unfortunately occurs, then the

team will lose $70,000. George, who is short of funds, will want Lloyd to

assume most of this loss. But, of course, Lloyd is not going to agree with

this unless his own shares are sufficiently high for the states C, D, and E.

He also might want George to share in some of the losses if A or B occurs,

just to keep George honest—or, more felicitously put, to give George the

right incentives.

George and Lloyd have to decide how to share the loss of $70,000 if

state A occurs. In order to keep our notation symmetric (which makes it

easier to generalize to more than two risk sharers at a later stage), let's

suppose that George and Lloyd have to select two numbers: AG, the payoff

to George if A occurs, and AL, the payoff to Lloyd if A occurs (all payoffs

1
If they were to disagree on the financial consequences associated with a given state, then they could
decompose that state into two or more states with differing probabilities. Our present format (including
additional states) is thus quite general. For example, if George thinks the payoff in state C is $30,000 and
Lloyd thinks it is $50,000, then state C could be split into two states, C' and C’’, with payoffs $30,000 and
$50,000, respectively. George may assign probabilities of .3 and zero to C' and C’’, whereas Lloyd may
assign probabilities of zero and .5 to C' and C’’, respectively.

Copyright  2003 Howard Raiffa 3

Ch 14 Addendum On Risk Sharing

are in thousands of dollars). We then require that AG + AL = -70. They have

to decide analogously on the splits in cases B. C, D, and E. SO overall,

George and Lloyd have to decide on ten numbers: AG, AL,…, EG, EL (see

Table 14A.1), subject to the following set of five constraints:

AG + AL = -70,
BG + B L = -20,
CG + CL = 30,
DG + DL = 80,
EG + E L = 200 .

For any determination of these ten numbers, George and Lloyd will

each be confronted with a risk profile. George's risk profile will yield

financial prizes AG, BG, . . . , EG with probabilities .07, .13, . . . , 10,

respectively; Lloyd's risk profile (see chapter 2) will yield financial prizes

AL, BL, . . . , EL with probabilities .05, .20, . . . , 05, respectively. Their

reactions to these lotteries will depend on their attitudes toward risk

taking. It could be that a specific risk-sharing plan (determined by a

specific setting of the ten numbers)2 is inefficient in the sense that the ten

risk-sharing numbers could be changed to improve the risk profile for each

party (in that party's subjective opinion). In other words, there may be

2
Because of the five financial constraints, these ten numbers have really five degrees of freedom: once we
determine what Lloyd gets in each state, George gets the complement.

Copyright  2003 Howard Raiffa 4

Ch 14 Addendum On Risk Sharing

opportunities for joint gains. Figure 14A.2 depicts graphically what could

occur. For a specific risk sharing plan, Q (which arises through the

specification of ten legitimate numbers), George might assign a certainty

equivalent to his resulting risk profile of $5,000, and Lloyd might assign a

certainty equivalent to his resulting risk profile of $13,000. However, as

depicted, the risk-sharing plan Q is not efficient: they both can improve,

since there are joint evaluations of risk-sharing deals that fall northeast of

Q. George controls the ownership of the deal and can remind (threaten?)

Figure 14.2. Set of joint evaluations of risk-sharing agreements.

Copyright  2003 Howard Raiffa 5

Ch 14 Addendum On Risk Sharing

Lloyd that there are other speculators who would love to join in the

venture. Lloyd could counter that he, too, has a choice of other potential

drilling deals. They also can remind each other about the transaction costs

of starting negotiations with other partners and how nice it would be to

work on other deals together in the future. The point of all this is that

Lloyd and George are involved in a negotiation that bears strong

similarities to other negotiations we have considered. This is not the place

to discuss details about how such sharing procedures are made or could be

made more efficiently. As in most negotiation processes, the protagonists

have to worry about their alternatives if they find it impossible to come to

an agreement. Each must consider the other external opportunities

available to him before he can arrive at a reservation price for the present

set of negotiations.

Suppose that George will deal with Lloyd only if he can get a

certainty equivalent of at least $15,000 from a mutually agreed upon risk

sharing deal; in other words, George's reservation price is $15,000. As

shown in Figure 14A.2, it may be possible to satisfy George and still get a

positive return for Lloyd, but there's not much leeway. They may never

find sharing arrangements that are mutually acceptable, even though such

agreements might exist.

Copyright  2003 Howard Raiffa 6

Ch 14 Addendum On Risk Sharing

The more structured the market, the easier it becomes to assess

objectively these externally driven reservation prices. We do not engage in

much haggling with our insurance companies when we obtain collision

insurance for our automobiles. But the owner of an oil tanker entering into

troubled waters might have some negotiating leverage with his insurance

suppliers, and vice versa. By and large, we can think of these negotiations

as collaborative decision making.

Copyright  2003 Howard Raiffa 7

Supplement to Negotiation Analysis: The Science and Art of Collaborative Decision Making by Howard
Raiffa (with contributing authors David Metcalfe and John Richardson), Harvard University Press, 2003.

Strictly Opposing Two-Player Games

by Howard Raiffa

Strategy of Presentation

We don’t seek generality but present enough of the theory in a special case to
enable the reader to generalize to more general cases.

We’ll continue with the development in Chapter 4 of Negotiation Analysis.

Consider the problem of two protagonists: Rowena and Colin who are strictly opposing.
If Rowena prefers A to B then Colin prefers B to A.

Let Rowena have two pure strategies R1 and R2 and Colin have strategies C1 and
C2 as was the case for the bi-matrix games of Ch. 4. For any (Ri, Cj) pair we need only
display Rowena’s utility payoffs since Colin’s motivation is to minimize Rowena’s
payoff from the game.

To emphasize the point that the payoffs are in utility units and that utilities are
essentially breakeven probabilistic judgments we’ll interpret the payoffs as follows: the
payoff to Rowena for the pair (Ri, Cj) is a utility number Aij between 0.0 and 1.0 which
gives Rowena an Aij standard probability at some desirable prize W and a
complementary probability at the status quo. Think of Colin supplying this prize to
Rowena so that the higher the number Aij the better for Rowena and the worse for Colin.

The initial case we examine has the structure

C1 C2

R1 A11 A12

R2 A21 A22

Rowena must choose either strategy R1 or R2; Colin has the choice of C1 or C2. The four
Aij numbers are known fully by the players; in the parlance of game theory, the payoff
numbers are common knowledge.

Copyright  2003 Howard Raiffa

Strictly Opposing Two-Player Games

We consider two qualitatively different games:

Table 1 Table 2

Game 1 Security Game 2A Security

Level Level
C1* C2 C1 C2

R1* .4 .5 .4* R1 .1 .9 .1

R2 .3 .2 .2 R2 .7 .2 .2*

Sec. Maximin Sec. .7* .9 Maximin

Loss .4* .5 = Loss <
Minimax Minimax

Analysis of Game 1

In Game 1 there is an equilibrium pair (R1, C1): If Rowena holds fixed at R1

there is no motivation for Colin to move from C1 to C2; and if Colin holds fixed at C1
there is no motivation for Rowena to move from R1. No other pair is in equilibrium. In
Game 1, by playing R1 Rowena can guarantee herself a return of .4 units. By playing C1
Colin can guarantee that Rowena can get at most .4 units. No other pair is in equilibrium.
Also note that the .4 entry is the max of its column and the min of its row.

In Game 1, Rowena, by playing R1, has a security level of .4; that .4 is the minimum of
row R1. By playing R2 Rowena has a security level of .2 which is the minimum of row
R2. The two row minima are .4 and .2 and the maxima of these two row minima is .4.
Turning to Colin’s viewpoint: the maximum that Colin can lose playing C1 is .4 (the max
of column C1; and the max he can lose with C2 is .5. Thus .4 and .5 are Colin’s security
losses and if he minimizes his security losses (i.e., if he minimizes his maximum
potential losses of .4 and .5 , he’ll choose C1 which guarantees maximum loss .4 units.

Game 1 has a value of .4 and Rowena by playing the strategy R1 that maximizes
the row minima – her so-called maximin strategy – can guarantee herself at least .4; and
Colin, by playing the strategy C1 that minimizes the column maxima – his so-called
minimax strategy – can guarantee that Rowena gets at most .4. Nice neat little reinforcing
bundle! How much different things are going to be when we turn our attention to Game
2.

Copyright  2003 Howard Raiffa 2

Strictly Opposing Two-Player Games

Analysis of Game 2

Check the following observations:

• There are no equilibria pairs in pure strategies. No payoff entry for
Rowena is both the maximum of its column and the minimum of its row.

• The minima in each row (labeled security levels) are .1 and .2 and the
maximum of these minima is .2. Thus the maximin value is .2 and the
maximin choice for Rowena is R2.

The maxima in each column (labeled security losses) are .7 and .9 and the
minimum of these maxima is .7. Thus the minimax value is .7 and the
minimax choice for Colin is C1.

• The maximin value of .2 – the value that Rowena can secure for herself –
is less than the minimax valueof .7 – the value that Colin can secure for
himself.

Figure 1:
___________________________________________
0 .2 .7 1.0

maximin minimax

gap

Randomized (or Mixed ) Strategies

Suppose that Rowena, instead of choosing a pure strategy of R1 or R2, tosses a

coin: heads she chooses R1 and tails R2. This is a so-called randomized (or mixed)
strategy which chooses R1 with standard probability of x = .5 and complementary
probability of 1 – x or also .5. Shortly we’ll consider the class of mixed strategies for x
values between.

If Rowena uses Mixed Strategy with x = .5, which we label MS(.5), then her
payoff against C1 is .5 x .1 + .5 x .7 = .4 ; and her payoff against C2 is .5 x .9 + .5 x .2
= .55. Note – and this deserves some thought – that all probabilities are standard or what
we have called canonical. We have studiously avoided, thus far, in any use of subjective
probabilities.

Copyright  2003 Howard Raiffa 3

Strictly Opposing Two-Player Games

Table 3

Game 2B

Security

C1 C2 Level

R1 .1 .9 .1

R2 .7 .2 .2

Toss Coin
MS(.5) .4 .55 .55*

Security Loss .7* .9

By introducing MS(.5) we have raised Rowena’s maximin value ( her security

level) from .2 to .55 and partially closed the gap between the minimax and the maximin
values.

We now have some work cut out for ourselves and some conjectures to be made:
• We have introduced for Rowena a new class of mixed strategies – the
MS(x) strategies. We now can manipulate x to achieve the optimal
maximin value.
• We can do the same trick for Colin: let him use C1 with probability y and
C2 with probability 1- y trying to improve his security losses. We should
suspect that by randomization he could improve (i.e., lower) his minimax
value.
• Wouldn’t it be “loverly “ if the gap between minimax and maximin values
vanished with proper choices of x and y. In this eventuality, the nice story
for game 1 would be repeated for Game 2 once mixed strategies are
introduced. There will be some value, v*, such that Rowena, by choice of
x*, can guarantee herself at least v* and Colin, by choice of y* can
guarantee that Rowena get at most v*. Furthermore the optimal x – let’s
call it x* – and the optimal y* are in equilibrium. We’ll see that’s the case.

Copyright  2003 Howard Raiffa 4

Strictly Opposing Two-Player Games

Geometry

Now to explain Fig. 1. The horizontal axis depicts the value of x, the probability of the
mixed strategy choosing R1. Hence as x goes from 0.0 (choice of R2 ) to 1.0
(choice of R1) the payoff against C1 goes from .7 on the left to .1 on the right; and
against C2 the payoff goes for .9 on the right to .2 on the left.

Now suppose Rowena were to announce she is going to use MS(.5) – i.e., she is
going to toss a fair coin. So we go to .5 on the horizontal axis and Colin has the choice
the C1 or the C2 lines. As the minimizing player, Colin will choose the C1 line against
MS(.5).But against x = .2 (say), Colin would prefer C2. Hence the broken darkened line
represents the payoff to Rowena as a function of x assuming that Colin chooses the best
retort from him. Thus the darkened line is the minimum function (or the security
function) that Rowena can achieve as a function of her chosen x. We see, and a little
algebra shows that Rowena should choose x* = 5/13 for her maximin value that will
guarantee her a return of 4.692. These numbers can be read off a precisely drawn figure
but they also can be derived algebraically as follows: We require

.1 x* + .7 ( 1- x*) = .9 x* + .2 (1 - x*)

and solving, we get x* = 5/13 = .385. The maximin value v*, is thus

v* = .1 (5/13 ) + .7 (8/13 ) = 6.1/13 = .4692.

End of story for Romena’s analysis and now on for more about Colin’s analysis.

1.0 1.0
.9 C2: y = 0 .9
.8 C1 : y = 1 Pivot Point
.8
.7 .7
.6 .6
.4692 .4 .4
.3 .3
.2 .2
.1 .1
0.0 0.0

0.0 .1 .2 .3 .4 .5 .6. .7 .8 .9 1.0

5/13 x values : probability of R1

Figure 2:

Copyright  2003 Howard Raiffa 5

 Strictly Opposing Two-Player Games

Colin’s Analysis

We could, but we won’t, present a figure for Colin analogous to Rowena’s.

Instead of the broken linear minimum function, Colin would be presented with a
maximum function and he would choose the minimum of the maximum function. Instead
we use Rowena’s geometry to develop Colin’s analysis.

Let y designate the probability that Colin uses C1, letting 1 – y be the probability
of C2. Now here is the key: for any specific y Colin uses, the return to Rowena can be
shown as a line in Fig.1 that goes through the pivot point. Indeed the lines for y = 0 and
for y = 1 are already shown on that diagram. Let’s examine the line we would get for
y = .5. It would be midway between lines for y = 0 and y= 1 but not quite horizontal. For
y = .5 the intercept on the left (the R2 side) would be .5 x .7 + .5 x . 2 = .45; and the right
intercept (the R1 side) would be at .5 x .1 + .5 x .9 = .5. Hence the line for y =.5 would
start on the left at a height of .45 and tilt slightly upwards , passing through the pivot
point and reaching a right of .5 at the right. If Colin were to announce y = .5, Rowena
would choose the right end point (corresponding to x =1 or to R1). Colin would like to
choose y such that the line it offers to Rowena is horizontal going through the pivot point
(as all lines for different y values do). To find this y we set

.1 y + .9 (1 – y) = .7 y + .2 ( 1 – y)
or
y = 7/13 = .538

and the height of the line is

.1 (7/13) + .9 (6/13) = 6.1/13 = .4692

The following display is helpful in keeping these calculated numbers in mind:

Table 4

Colin’s minimax strategy

Rowena’s 7/13 6/13
maximin
strategy 5/13 .1 .9 6.1/13

8/13 .7 .2 6.1/13

6.1/13 6.1/13

The calculations say: Rowena can guarantee a payoff of 6.1/13 by playing her maximin
strategy. Colin can guarantee that Rowena gets at most 6.1/13 by playing his minimax

Copyright  2003 Howard Raiffa 6

Strictly Opposing Two-Player Games

strategy. Rowena’s maximin and Colin’s minimax strategies are in equilibrium in the
sense that there is no motivation for either to change if the other holds fixed. End of
story.

Extensions to Bigger Games

[Use of an EXCEL Linear programming package to find Rowena’s maximin strategy and
Colin’s minimax strategies and the resulting value of the game with more than two
strategies for each player.]

Let m and n represent respectively the number of pure strategies for Rowena and
Colin. We have just completed the story for m = n = 2. We now investigate the analysis
for more general m and n.

First let us illustrate a game that has an equilibrium pair in pure strategies. See
Table 5. We want to continue interpreting the payoff numbers as utilities so think of the
payoffs as chances out of ten that Rowena will receive a desirable prize from Colin.

Table 5
C1 C2 C3 C4 Security Level

R1 1 3 4 1 1

R2 6 5 8 7 5 maximin

R3 4 4 0 2 0

Security Loss 6 5 8 7
minimax

In the game of Table 5, m = 3 and n = 4. We assume the players have complete

knowledge of the structure and payoffs; and each knows the other knows etcetera. That’s
what we mean by common knowledge.

Rowena’s security levels for her pure strategies are the minima of the rows and
the maximizer of these row minima is R2. Her maximin value in pure strategies is 5. She
can guarantee at least 5 units for herself by playing R2. Colin’s security losses are the
maxima of each of the columns and the minimizer of these column maxima is C2; he can
guarantee, by playing C2 that Rowena gets no more than 5 units. In this case the maximin
equals the minimax.

Copyright  2003 Howard Raiffa 7

Strictly Opposing Two-Player Games

The pair (R2, C2) is in equilibrium since R2 is best for Rowena against C2; and C2 is
best for Colin ( the minimizing player) against R2. Observe also that the (R2, C2) entry
of 5 units is the maximum of its column and the minimum of its row. This summarizes
what need be said about strictly competitive games with pure strategy equilibria.

Games with no equilibrium pair in pure strategies. Mixed strategies to the rescue.

Consider the game depicted in Table 6. The row minima are 1, 1, 2 and the
maximum of these minima is 2, Rowena’s maximin value in pure strategies. The maxima
in each Column are respectively 5, 3, 4, 6 and the minimum of these column maxima is
3.Thus Colin can secure, by choice of C2, not to lose more than 3 units but Rowena can
only command 2 units for herself. There is a gap between the maximin value and the
minimax value. We also note that there is no equilibrium pair – no entry is both max in
its column and min in its row.

Table 6 No Equilib rium Pair

Row
C1 C2 C3 C4 Minima
R1 1 3 4 1 1
R2 5 2 1 6 1
R3 4 3 3 2 2 Maximin

Column Maxima 5 3 4 6
Minimax

We will now show that Rowena can increase her security level – i.e., increase her
maximin value by use of randomized strategies.

Consider for example the randomized strategy that assigns equally likely values to
R1, R2, and R3 – i.e., assigns a .333 prob to each pure strategy as shown in Table 5.

Copyright  2003 Howard Raiffa 8

Strictly Opposing Two-Player Games

Table 7
Randomized Row
Strategy C1 C2 C3 C4 Security
0.333 R1 1 3 4 1 1
0.333 R2 5 2 1 6 1
0.333 R3 4 3 3 2 2

random strategy 3.333 2.667 2.667 3.000 2.667

If Rowena uses this randomization, then her expected return against any column chosen
by Colin is the simple average of the entries in that column. Hence her payoffs in utility
terms are as shown in the last row of Table 7 and the security level of that randomized
(or mixed) strategy is the minimum of those entries, a whopping 2.667 value. Can she do
better by using a different randomization? How much can she aspire to? Certainly not
above Colin’s minimax value of 3? Can she aspire to get a secutity level of 2.9? This
becomes now a mathematical programming problem. First let us note that the value for
Rowena’s entry against C1, which we already know is the average of the C1 column of
numbers, is also the sumproduct of the strategy column and the C1 column. Indeed the
payoff of any legitimate randomization against any column is merely the sumproduct of
those two columns of numbers.

Use of EXCEL to Compute Rowena’s Highest Security Level Using Mixed

Strategies.

We first set up Table 8. We choose a ballpark aspiration level of 2.5. That number
is used as a convenience to set the calculations and is arbitrary at this time. Our aim is to
find a legitimate randomized strategy – a set of three non-negative numbers that sum to
1.0 – that will yield expected payoffs against C1 to C4 that are each in excess of some
pre-specified aspiration level. In this case we see that the randomized strategy as shown
yield numbers that are each in excess of the aspiration value of 2.5. These excesses are
shown in the last row of numbers in Table 6. The problem is: Given the setup in Table 8,
maximize the aspiration-level cell and find a set of legitimate randomization values for
which each of the excesses is non-negative. This is a linear programming problem ideally
setup to use SOLVER, an attachment on EXCEL.

Copyright  2003 Howard Raiffa 9

Strictly Opposing Two-Player Games

Table 8 Aspiration 2.5

Randomized
Strategy C1 C2 C3 C4
0.333 R1 1 3 4 1
0.333 R2 5 2 1 6
0.333 R3 4 3 3 2

Random strategy 3.333 2.667 2.667 3.000

Excess 0.833 0.167 0.167 0.500

Sumproduct of randomized strategy

column and C1 column

When SOLVER was queried it responded by showing Table 9, which shows how
Rowena could choose a randomized strategy yielding her maximum possible aspiration
of 2.75.

Table 9 Aspiration 2.75

Randomized
Strategy C1 C2 C3 C4
0.250 R1 1 3 4 1
0.250 R2 5 2 1 6
0.500 R3 4 3 3 2
1.000

random strategy 3.500 2.750 2.750 2.750

Excess 0.750 0.000 0.000 0.000

sumproduct of randomized strategy

column and C1 column

Colin’s Analysis
We start the process in Table 11. We choose a representative random strategy for
Colin as shown in B3:E3. The sum of those four probabilities is shown in cell F3, which
in this case is 1.0. The expected payoffs of this randomized strategy against R1 is 2.3;

Copyright  2003 Howard Raiffa 10

Strictly Opposing Two-Player Games

against R2 is 3.6; against R3 is 2.7. These are sumproducts of the randomized strategy
and the row payoffs. With a representative arbitrary aspiration level of 2.9 (pulled out of
the air) we see that Colin cannot achieve this aspiration level using the displayed random
strategy (B3:E3) in because the excess with R2 I s negative. Remember that Colin is the
minimizing player and the excesses are the aspiration value minus the EV columns. We
see that as pitted against R2 the resulting EV of 3.6 exceeds the aspiration level of 2.9.

A B C D E F G
1 Table 11 Aspiration 2.9
2
3 C’s Rand. Str. 0.1 0.2 0.3 0.4 1
4 C1 C2 C3 C4 EV Excess
5 R1 1 3 4 1 2.3 0.6
6 R2 5 2 1 6 3.6 -0.7
7 R3 4 3 3 2 2.7 0.2

Colin’s Linear Programming Problem:

Given the payoff table (in B5:E7);
find
legitimate values for the four adjustable numbers in B3:E3
– – i.e., numbers that are non- negative and whose sum
appearing in cell F3 is 1.0
and an aspiration level in D1 ;

to maximize
the entry in D1

subject to
keeping the calculated values ( the excesses) in F5:F7 non-negative.

SOLVER takes a second or so to produce the answer shown in Table 12. Colin can hold
Rowena down to at most 2.75 by playing his minimax randomized strategy.

Copyright  2003 Howard Raiffa 11

Strictly Opposing Two-Player Games

Table 12 Aspiration 2.75

Colin’s minimax Rand. Str. 0 0.5 0.25 0.25 1

C1 C2 C3 C4 EV Excess
R1 1 3 4 1 2.75 0
R2 5 2 1 6 2.75 0
R3 4 3 3 2 2.75 0

The dual solution – Rowena’s as well as Colin’s – appears in Table 13.

Table 13
A B C D E F G H
1
2
3 0 0.5 0.25 0.25 1 Row’s EV excess

4 0.25 1 3 4 1 2.75 0
5 0.25 5 2 1 6 2.75 0
6 0.5 4 3 3 2 2.75 0
7 1
8 Col’s EV 3.5 2.75 2.75 2.75 2.75

That completes the story except for a proof that argues that the best aspirations for
Colin and Rowena are always the same. The gap with randomization between the
maximin and minimax values is zero. We know how to make this result convincing but
it’s too specialized for our needs and is thus omitted.

Copyright  2003 Howard Raiffa 12

Supplement to Negotiation Analysis; the Science and Art of Collaborative Decision Making by Howard
Raiffa (with contributing authors David Metcalfe and John Richardson), Harvard University Press, 2003.

Strictly Opposing Two-Player Games:

Security Analysis
Fig. 1 : (R2,C2) in
equilibrium

C1 C2 C3 C4
R1 1 3 4 1
R2 6 5 8 7
R3 4 4 0 2

Max of
Col. and
Min of
Row

Fig 2 :
Row
C1 C2 C3 C4 Security
R1 1 3 4 1 1
R2 5 2 1 6 1
R3 4 3 3 2 2 Maximin

Column 5 3 4 6
Security
Minimax

Fig. 3
Row
C1 C2 C3 C4 Security
R1 1 3 4 1 1
R2 5 2 1 6 1
R3 4 3 3 2 2 Maximin

Column 5 3 4 6
Security
Minimax

No equil. pair in
pure strategies

Copyright  2003 Howard Raiffa

Strictly Opposing Two-Player Games: Security Analysis

Fig. 4 : Intro of
Randomized
str.
Strategy C1 C2 C3 C4
x1 R1 1 3 4 1
x2 R2 5 2 1 6
x3 R3 4 3 3 2
1 EV
Sum
products
of Xs and
Cs

Fig. 5:

Table 6 Aspiration 2.75

Randomized Row
Strategy C1 C2 C3 C4 Security
0.250 R1 1 3 4 1 1
0.250 R2 5 2 1 6 1
0.500 R3 4 3 3 2 2
1.000
random 3.500 2.750 2.750 2.750
strategy
Excess 0.750 0.000 0.000 0.000

sumproduct of column and

randomized C1 column
strategy

How far can we

push up the
aspiration of
2.5 and still
keep all
excesses
positive?

Copyright  2003 Howard Raiffa 2

Strictly Opposing Two-Player Games: Security Analysis

Aspiration 2.75

Randomized
Strategy C1 C2 C3 C4
0.250 R1 1 3 4 1
0.250 R2 5 2 1 6
0.500 R3 4 3 3 2

Randomized Strategy 3.500 2.750 2.750 2.750

Excess 0.750 0.000 0.000 0.000

Table 9 Aspiration 2.75

0 0.5 0.25 0.25 1

C1 C2 C3 C4 EV Excess
R1 1 3 4 1 2.75 0
R2 5 2 1 6 2.75 0
R3 4 3 3 2 2.75 0

Table 9 Aspiration 2.75

Rand. Str. 0 0.5 0.25 0.25 1

C1 C2 C3 C4 EV Excess
R1 1 3 4 1 2.75 0
R2 5 2 1 6 2.75 0
R3 4 3 3 2 2.75 0

Copyright  2003 Howard Raiffa 3

Strictly Opposing Two-Player Games: Security Analysis

A B C D E F G
1 Table 11 Aspiration 2.9
2
3 Rand. Str. 0.1 0.2 0.3 0.4 1
4 C1 C2 C3 C4 EV Excess
5 R1 1 3 4 1 2.3 0.6
6 R2 5 2 1 6 3.6 -0.7
7 R3 4 3 3 2 2.7 0.2

A B C D E F G H
1
2 rand.
3 0 0.5 0.25 0.25 1 str. excess
4 0.25 1 3 4 1 2.75 0
5 0.25 5 2 1 6 2.75 0
6 0.5 4 3 3 2 2.75 0
7 1
8 random str 3.5 2.75 2.75 2.75 2.75
9 excess 0.75 0 0 0
10

A B C D E F G H
1
2 rand.
3 0.1 0.3 0.4 0.2 1 str. excess
4 0.1 1 3 4 1 2.8 -0.3
5 0.5 5 2 1 6 2.7 -0.2
6 0.4 4 3 3 2 2.9 -0.4
7 1
8 random 4.2 2.5 2.1 3.9 2.5
str
9 excess 1.7 0 -0.4 1.4

Copyright  2003 Howard Raiffa 4

Strictly Opposing Two-Player Games: Security Analysis

A B C D E F G H
1
2 rand.
3 0 0.5 0.25 0.25 1 str. excess
4 0.25 1 3 4 1 2.75 0
5 0.25 5 2 1 6 2.75 0
6 0.5 4 3 3 2 2.75 0
7 1
8 random 3.5 2.75 2.75 2.75 2.75
str
9 excess 0.75 0 0 0

Copyright  2003 Howard Raiffa 5